Don’t take my word for it. Twice, researchers tested the same drug. Against the same condition. They got the same result, and they published opposite conclusions. One was called significant. One was called no effect. What gives? If you guessed that the “significant” study simply used a bigger sample size, congratulations. That is the entire difference.

This story comes from a comment in Nature in 2019. More than 800 scientists signed it. The same comment collects surveys of 791 published articles from five journals, in fields from conservation biology to psychology to sociology. Care to guess how many of those authors read a non-significant result as proof of no effect whatsoever? About half.

The high-impact journals do it too. A separate team checked 513 papers in five top journals: Science, Nature, Nature Neuroscience and two others. Those papers made 157 direct comparisons between significant and non-significant results. Only 78 of the 157 were done correctly. 79 were not. Flip a coin, because suddenly science doesn’t look very trustworthy at all.

And we were warned. In 2016, the American Statistical Association issued an official statement advising against basing scientific conclusions on whether a p-value crosses a threshold. We do it anyway.

Two studies found exactly the same risk ratio of 1.2. One was called significant. The other was not.

Read the numbers in that image. Both studies found a risk ratio of 1.2. Patients taking the drug had a 20 percent higher risk of the side effect. Study 1 found that 1.2 with a 95 percent confidence interval of 1.09 to 1.33. Its p value was 0.0003. Bingo. Significant. Study 2 also found 1.2. But its sample was smaller, so its interval was wider: 0.97 to 1.48. Smaller sample sizes give less precise answers. Its p-value was 0.091. Nothing, apparently. Not significant. Scientists see the stars and forget the intervals. Look how much they overlap! These two studies barely disagree at all.

Where did 0.05 come from? Not from nature. Ronald Fisher wrote in 1926: “Personally, the writer prefers to set a low standard of significance at the 5 per cent point”. And that was it. One opinion, from one very influential person, repeated until it became common wisdom. Heck, you can now accept or reject papers with it. Show a reviewer that you calculated the statistics by putting a star next to your table.

Understanding what that star means? Much harder. As far as we can tell, hardly anyone manages it. Chinese scientists tested 1,479 of their colleagues and students on what p-values mean. Get this: 89 percent failed. Failed! Graduate students didn’t do better than professors. Statistics teachers didn’t do better than anyone else. Have you ever read a non-significant result as no effect? Then welcome to the majority. This isn’t advanced statistics confusing students. This is basic statistics that almost every researcher misreads. Including the ones who teach it.

Meanwhile, large samples quietly hand out stars for free. Facebook ran an experiment on 689,003 users. They measured effects as small as d = 0.001. That’s about as close to zero as you can get without being zero. Every effect they reported was statistically significant. With a big enough sample, almost anything is.

You can only use a p-value well if you remember which question it answers. Just one: if there were no effect in the population, what is the chance I would see data as extreme as mine, from noise alone? That’s it. That’s the whole service. For every other question you want to ask, your p-value cannot help you. Does my effect matter? Thats not a statistical question. That one belongs to biologists, or clinicians, or economists. And you already know where the answer lives. It’s written right there in your units. The American Statistical Association put it better back in 2016: a p-value “does not measure the size of an effect or the importance of a result”.

Here’s what to do instead. A p-value of 0.06 usually means you found a real effect but measured it imprecisely. Report it! Put the effect and its confidence interval in your paper. You might be helping the meta-analyst who will one day combine your study with a more precise one. A p of 0.001 attached to a tiny effect means you measured something tiny, very precisely. Say so. And report your effect in units that matter to your question. Milligrams. Dollars. Hectares. Then tell your reader what the world would look like at either end of your interval. Imagine your story if the true effect sits at the top of it. Now imagine it at the bottom. What would a meaningful difference look like in your experiment? Tell your reader that. We can disagree about whether your effect is important. Nobody can disagree with milligrams, or hectares, or dollars.

If you are still working on your first paper, my book covers how to write results that editors trust. It is called How to Publish Your First Research Paper, and it is on Amazon: https://www.amazon.com/dp/B0H4RM4NXD

Sources. The drug example and the survey of 791 articles are from Amrhein, Greenland and McShane, Nature, 2019. The neuroscience count is from Nieuwenhuis, Forstmann and Wagenmakers, Nature Neuroscience, 2011. The survey of 1,479 researchers is from Lyu and colleagues, Journal of Pacific Rim Psychology, 2020. The Facebook experiment is from Kramer, Guillory and Hancock, PNAS, 2014. Fisher’s sentence is from The Arrangement of Field Experiments, 1926. The p value definition is principle five of the American Statistical Association statement, 2016.

I want to start with an editor. I have invented her. But everything she does in the next few minutes comes from published figures, and you can check every one at the bottom of this issue. I have also been an editor, so I say this partly from experience.

It is a Tuesday morning in March. She handles about forty papers at a time for a respectable journal in your field, unpaid, on top of her own teaching and her own lab.

Your paper is fourth in her queue. She opens it at twenty past seven, before her first meeting.

She reads your abstract, then your figures, then your abstract again. She likes it. It is not perfect, and there is a control she would have run differently, but it is careful work and it belongs in her journal. So she does not reject it.

Now she needs two people to read it.

She opens her list. The obvious first choice is a professor in Utrecht who wrote the key paper on your method. The second is a group leader in São Paulo. Then a postdoc in Melbourne, and an associate professor in Nairobi who reviewed beautifully for her last year. She writes the four invitations herself, because the automatic ones get ignored, and sends them at four minutes to eight.

Then she waits.

And so do you.

How this is supposed to work

Peer review is one of the genuinely fair things left in academic life.

You cannot buy your way past it. It does not matter who your supervisor was, or which country your university sits in, or whether anybody outside your department has heard of you. Your paper goes to people who work on what you work on. They read the argument and they check the evidence, and either it holds up or it does not. If it holds up, it is published, and from that day it counts exactly as much as anybody else’s paper.

The machinery is simple enough. You submit. An editor decides whether the paper is worth sending out. If it is, two or three specialists read it and write reports. The editor weighs those reports, makes a decision, and sends it to you. You revise, and it goes round again.

Every field runs some version of this. Medicine, physics, economics, history. It is slow, and it is unpaid, and it is the best system anybody has come up with.

What you think is happening

So you submit in March, and nothing comes back.

April passes. May passes. The status line says “Under Review” and does not change. By the middle of June you are doing arithmetic in your head. Your contract, your visa, your next application, and the paper you cannot cite yet.

And the story you tell yourself is a reasonable one. Two experts have your paper. They are reading it slowly, the way careful people read. Perhaps they disagree with each other. Perhaps the editor has asked for a third opinion. The word review means somebody is reviewing, so somebody is reviewing, and four months is simply what care costs.

That is what I assumed for most of my career.

It is not what is happening.

Where the four months actually go

Reviewers are not slow.

PLOS ONE gives its reviewers 10 days to file a report. Frontiers gives 10. MDPI gives 14. And once somebody has agreed to read your paper, the middle time they take to finish is 16.4 days, measured across every field.

Two or three weeks. That is the reading. So where are the other eleven weeks?

Table 1: where the months go

The numbers: Reviewer’s deadline: 10 to 14 days. Time a reviewer takes once they agree: 16.4 days. Invitations an editor must send to get one finished review: 1.9 in 2013, rising to 2.4 in 2017. Reviewers saying yes at one long-running journal: above 60% in 2003, just under 40% in 2023.

Now go back to the Tuesday morning in March.

Utrecht declined in two days, politely, citing a grant deadline. São Paulo never replied at all. Melbourne accepted, then stopped answering email. Nairobi accepted and filed a thoughtful report in eleven days, which is the whole reason this system still functions.

It is now the second week of May. Your editor has one report, needs two, and has started again from the top of her list.

Laura Meyerson and her colleagues went through twenty-one years of invitations at the journal Biological Invasions. One sentence in their paper explains your four months better than anything else I have read on the subject:

“Editors are often still looking 60 or 90 days after submission for qualified reviewers who will agree to assess the manuscript.”

Sixty to ninety days. Looking. Not reading or weighing your results, not arguing with a co-editor. Looking for a person.

Nobody had opened your figures. Nobody disliked your controls. Your paper had not failed anything, because nothing had happened to it yet.

One more line from that study, which I find hard to read. Early-career researchers said yes to review invitations 48.5% of the time. Senior professors said yes 33.5%. The people with the least time and the least security are carrying the most of this.

So what should you expect

This is the best cross-field picture we have, and the evidence is surprisingly thin given how many careers it decides. It comes from a site called SciRev, where researchers log what happened to their own submissions. Jeroen Huisman and Jeroen Smits went through about 3,500 of those reports.

Table 2: how long peer review takes, by field

The numbers: Average weeks for the first review round, then the share answered within three months, then the total time under review across all rounds.

A medical paper gets an answer in about 8 weeks, an economics paper in about 18. Advice that travels from one of those fields to the next valley, let alone the next continent, is worth very little, and most of the advice online is exactly that.

One piece of good news is buried in that table. It leaves out papers turned away without review, which are counted separately and come back quickly, at 16 days on average. So if you are three or four weeks in and have heard nothing, your paper has probably cleared the first gate.

And in 2015 Vivian Nguyen and colleagues asked 461 authors two separate questions. How long should peer review take, and how long did it actually take you? The authors said the review felt slow at about 14.4 weeks. The wait they reported having was 14.4 weeks. The point at which you start to feel that this has gone on too long is, on average, exactly what you get.

The number on the journal’s website will not help you

This is the thing I would change first, if anyone asked me.

Open the Nature journal metrics page. It says submission to first editorial decision: 8 days. Read one line further down the same page. Submission to acceptance: 321 days.

Both are true, and Nature is not hiding anything. The definition is printed right there: the 8 days is the time until a paper is “either sent out for peer review or rejected.” It measures how long you wait to learn whether your paper goes to reviewers at all. Nature turns away the large majority of what it receives, so the typical submission never reaches a reviewer. That 8 days describes the rejection queue rather than peer review.

Table 3: the advertised speed versus the real one

The numbers: Nature: 8 days to first decision, 321 days to acceptance, 2025 data. Nature Communications: 9 and 242. Nature Human Behaviour: 6 and 252. Nature Physics: 11 and 195.

Three Springer journals print the same kind of figure with no definition next to it. Statistics and Computing says 3 days. Synthese, a philosophy journal, says 4 days. Mathematische Annalen, one of the oldest pure mathematics journals in the world, says 14 days. Ask any mathematician whether a serious paper can be refereed in fourteen days. Proofs are read line by line, queried, and read again, and it takes months.

Taylor & Francis admits this in its own guidance, to its credit. Journals that reject a lot of papers without review show shorter times, it explains, “potentially as low as 0 days.”

A few journals do publish the split. The Economic Journal reported that in 2024 it rejected 43% of papers without review. Average turnaround for everything: 37 days. For the ones rejected without review: 6 days. For the ones that actually went to referees: 62 days.

So an undefined speed figure on a journal page should be read as a measure of how picky that journal is rather than how fast it reviews. It is not dishonest. It is just not the number you came for.

Four things to do instead of waiting

1. Check the journal’s real speed before you submit. Most serious journals are required to print the submission and acceptance dates on every paper. Open eight recent papers in the journal you are considering and do the arithmetic yourself. It takes twenty minutes, it is current, and it describes that one journal rather than an average of thousands. If you take one thing from this issue, take this.

2. Post a preprint, and post it before you submit. A 2024 study of 100,077 preprint papers found faster acceptance, but only when the preprint went up before the paper went to a journal. Posting afterwards gave no benefit. The middle gap between preprint and journal version is 155 days. That is five months where your work is public, dated, and usable on a job application instead of invisible.

3. Learn what your status line means, then time your email to it. In Editorial Manager, which most Elsevier and Springer journals use: With Editor means an editor has it and nothing more. Reviewer Invited is the hunt described above. Under Review means reviewers have agreed, so the hard part is behind you. Required Reviews Completedmeans every report is in and nobody outside the journal is holding things up. That last one is the moment to write. Three sentences, the manuscript number, and one question, which is whether there is an expected date. I have never known that to annoy an editor.

There is no three-month rule, by the way. I went and checked, because I half believed it myself. Elsevier, Springer Nature, Wiley and Taylor & Francis all decline to name a waiting period. Elsevier’s advice is to measure your wait against that journal’s own published figure rather than against a number somebody repeated in a forum.

If you are publishing from a lower-income country

Andy McKay edits the Review of Development Economics, and he did something I wish more editors would do. He published the numbers on his own journal. Authors from low and middle income countries sent in 61% of the papersand got 33% of the published ones.

The gap opened mostly at the desk stage. Papers from poorer countries were much less likely to be sent out for review at all. After review, the difference was, in his words, “much less stark.”

Language is a real cost, and it is measured rather than imagined. Among second-language English speakers at a middle level of English, 38.1% had had a paper rejected because of the writing. Among native speakers it was 14.4%.

I want to give you the other side too, because it changes what you should do about it. When researchers gave 4,147 scientists the same abstracts with the author addresses swapped around, they found almost nothing. Only 2 of 36 comparisons came out real, and one of those went the wrong way for the theory.

Put those together and you get something more useful than “the system is against you.” The gap seems to open through language, framing, and fit with the journal. It opens at the desk check, not when a reviewer glances at your address.

I think that is good news, for one reason. The desk stage is the part you can change, and it is settled in days rather than months. Does the paper fit this journal, is the cover letter clear, is the formatting right, and is the English clean. Get those four right and you are in the queue with everybody else, waiting on a Tuesday-morning editor who is doing her best.

What I would not do with these numbers

Do not compare across the tables. The by-field table gives averages, and a few very slow papers drag an average upwards. The journal figures are mostly middle values, which behave differently.

The by-field data is from 2013 to 2016, and it comes from volunteers. People who had a bad experience are likelier to file a report, and the authors say so themselves.

Every journal figure counts only the journal that finally published the paper. None of it includes the months you lost at the journals that turned you down first. The real time from finished draft to published paper is longer than anything here, and as far as I can tell nobody has measured it.

Sources

Time by field: Huisman & Smits (2017), Scientometrics 113(1), 633-650, via scirev.org. Journal figures: the 2025 metrics pages of Nature, Nature Communications, Nature Physics and Nature Human Behaviour; the Springer journal pages for Statistics and Computing, Synthese and Mathematische Annalen; Taylor & Francis journal metrics guidance; The Economic Journal 2024 Annual Report. Reviewer behaviour: Publons and Clarivate, 2018 Global State of Peer Review; Meyerson, Suzzi-Simmons & Simberloff (2025), Biological Invasions 27:223; the PLOS ONE and Frontiers peer review pages. What authors expect: Nguyen et al. (2015), PLOS ONE 10(8): e0132557. Statuses: the Cochrane Editorial Manager knowledge base and Elsevier author support. Preprints: Tian et al. (2024), Scientometrics 129(7); Fraser et al. (2020), eLife 9: e52646. Developing countries: McKay (2024), Review of Development Economics 28(4); Amano et al. (2023), PLOS Biology 21(7): e3002184; Nielsen et al. (2021), eLife 10: e64561.

I keep a longer document with a link to every one of these, the period each figure covers and how much I trust it. If you want it, just reply to this email and I will send it over.

Thank you

If this was useful, send it to the person in your lab who is refreshing a submission page right now and taking the silence personally.

And if you are still working on your first paper, I wrote a book about that part. It is called How to Publish Your First Research Paper: The Unwritten Rules of Getting Accepted in International Journals, and it is on Amazon for $6.99. It is the thirty years of advice I usually end up giving in my office, written down.

Good luck with the one you are waiting on.

— John

The full tables are below.

Why I built this

A few weeks ago a former student wrote to me with two offers in front of her. One was in Germany. One was in the Gulf. The German number was bigger. She had already decided to take it, and she was writing to tell me, not to ask me.

I asked her one question. Did she know what her rent would be?

She did not. Nobody had told her that the German offer was a 65% position, which is normal in some fields there. Nobody had told her that the Gulf offer came with an apartment. When we worked it out on a Sunday afternoon, the “smaller” offer was worth more money, more space, and more savings at the end of two years.

She took the other job.

That conversation happens more often than it should. Postdoc salaries are published in dozens of different formats, in dozens of currencies, and almost nobody puts them side by side. So I did. This issue is the result: 45 countries, one table, ranked two ways.

First, the number everyone quotes

Here is what a standard starting postdoc is paid, before tax, converted at market exchange rates on 22 August 2026.

The headline numbers, in plain text: Switzerland pays $109,871 a year. Austria pays $82,007. Belgium pays $78,738. Australia pays $75,432, and adds 17% employer pension on top of that. Germany pays $66,718 at a full-time E13 post. Norway pays $66,137. Qatar pays $65,934, the United Arab Emirates $65,351 and Saudi Arabia $64,000, all three with no income tax. Denmark pays $65,338. The United States pays $63,480 under the NIH standard rate, which is the floor most American universities build on.

At the other end of the same table: India pays about $10,026 a year, Turkey $10,864, Pakistan $3,555, Egypt $2,358 and Nigeria about $1,114.

That is a gap of almost 99 times between the top and the bottom. It is a shocking number, and it is also a misleading one.

Second, the number nobody quotes

A dollar does not buy the same thing in Zurich and in Chennai. So the fair question is not “how much does the job pay,” it is “how much does the pay buy.”

There is an official way to answer that. The World Bank publishes something called a purchasing power parity conversion factor for every country. In plain words: it measures how expensive normal life is in that country compared with the United States. Rent, food, transport, electricity, a haircut, a bus ticket. If prices are half as high, your money goes twice as far.

I used the 2025 figures and converted every salary above into what it would buy at United States prices. That is the “real value” column below.

Positions 1 to 23, in plain text: Saudi Arabia $128,266. United Arab Emirates $94,975. Austria $93,556. Belgium $87,336. Qatar $87,099. Switzerland $82,587. Germany $79,445. Hong Kong $72,836. Australia $71,892. Singapore $70,325. Poland $67,857. Netherlands $64,493. United States $63,480. South Korea $62,491. Norway $61,492. Italy $61,330. Finland $61,096. Denmark $60,376. New Zealand $60,306. China $57,811. Taiwan $57,202. Chile $57,066. United Kingdom $56,884.

Positions 24 to 45, in plain text: Canada $55,534. Ireland $51,778. Sweden $51,299. Spain $50,341. Vietnam $49,977. India $48,390. France $47,508. Malaysia $46,133. Thailand $45,256. Japan $42,037. South Africa $41,344. Portugal $39,693. Israel $37,749. Kenya $32,525. Czech Republic $32,502. Philippines $32,143. Turkey $29,418. Mexico $28,650. Brazil $24,172. Egypt $15,540. Pakistan $14,680. Nigeria $4,729.

Look at what happens to the gap. On paper, Switzerland pays 99 times what Nigeria pays. Adjusted for prices, Switzerland pays 17 times what Nigeria pays. The widest gap in the whole table, from Saudi Arabia at the top to Nigeria at the bottom, is 27 times.

Still enormous. But a different world from the first table.

Who moves, and by how much

This is my favourite table in the issue, because it shows which countries are quietly good and which ones are quietly bad.

The movers, in plain text. Poland climbs from 25th to 11th, a jump of 14 places, and is the best value in Europe that almost nobody talks about. India climbs 13 places, from 42nd to 29th. South Korea, Taiwan and Hong Kong each climb 10. Saudi Arabia climbs 9 places to first. China climbs 8.

Going the other way: Israel falls 14 places, from 22nd to 36th, because the fellowship is modest and the country is expensive. Denmark and Norway each fall 9 places. Ireland falls 9. France falls 7. Mexico falls 7. Sweden and the Czech Republic each fall 6.

Switzerland, for the record, falls from first to sixth. It still pays extremely well. It is just that Zurich charges you for the privilege.

The Gulf point, and I will be honest about my position

I work at the American University of Sharjah, so treat what follows as informed rather than neutral.

Three things are true about a Gulf postdoc offer at the same time, and they compound.

There is no income tax. In the United Arab Emirates, Saudi Arabia and Qatar, the number in the offer letter is close to the number that reaches your bank account. A German postdoc on €57,112 does not keep €57,112. A postdoc in Sharjah on AED 240,000 keeps roughly AED 240,000.

Housing is usually separate. The American University of Sharjah fellowship advertises up to AED 20,000 a month plus on-campus housing for single fellows, health insurance and a flight home each year. KAUST in Saudi Arabia provides on-campus accommodation and, on its Global Fellowship, publishes a $75,000 stipend with up to $40,000 of research money. Housing is normally the single biggest line in a young researcher’s budget. Removing it changes everything.

Prices are lower than the salary suggests. That is what pushes the Gulf to the top of the real-value table.

Now the honest caveats. These are fixed-term contracts with no path to permanent residency in most cases. The published salary scales are also thin. Gulf universities mostly do not publish postdoc pay tables. So some of my Gulf figures come from job adverts and reported salaries, not from an official scale. And a two-year contract that ends with no local job market is a real risk, whatever it paid. I am not telling you the Gulf is the right answer. I am telling you it belongs in your comparison, and it usually is not in it.

The trap that catches most people

Two offers with the same number can be worth very different amounts. Here is what actually sits underneath the headline figure.

In plain text, the things that change the answer most:

Tax. Germany takes roughly 35% to 40%. Denmark takes roughly 38% to 45%. The Gulf states take nothing. Turkey’s TÜBİTAK fellowship, Portugal’s FCT stipend, Brazil’s CNPq scholarship, Mexico’s SECIHTI grant, South Africa’s NRF fellowship and Israel’s postdoc fellowship are all tax-free too.

But tax-free is not free. In Brazil, Mexico, South Africa, Israel and Portugal, “tax-free” means you are not an employee. No pension is being built for you. No unemployment insurance. No labour law protects you. South Africa’s NRF fellowship pays ZAR 320,000 a year with no pension, no maternity leave and no employment contract, and South African postdocs have complained about this for years. Mexico’s official 2026 call states in writing that the award creates no employment relationship.

Pension you cannot see. Australia adds 17% employer superannuation on top of the salary. Denmark adds 17.1%. Neither appears in the advertised number, and both are real money.

Part-time posts advertised as full-time careers. In Germany, many positions, especially in the humanities and social sciences, are funded at 65% of a full post. That turns €57,112 into €37,123. The job title does not change. The workload often does not either.

Stipend versus salary. Japan’s JSPS fellowship pays ¥362,000 a month, which is the best-paid standard route in Japan, but it is a stipend with no employer social insurance. You pay that yourself, so your take-home is worse than a salaried post of the same size. The same distinction applies in the United States between an NIH training stipend and a normal university appointment.

Where the numbers are genuinely bad

I want to be careful here, because the “your money goes further at home” argument can be used to excuse very low pay.

It does not work everywhere. Nigeria comes last on both lists. Even after adjusting for prices, a Nigerian academic’s basic pay is worth about $4,729 a year at United States prices. Nigeria has no national postdoctoral scheme at all. That figure is the Lecturer II basic salary a new PhD is actually appointed to, after the 40% increase agreed in January 2026. Pakistan comes in at $14,680 and also has no standard domestic postdoc scale. Egypt is at $15,540, and the Egyptian pound lost roughly half its value against the dollar across 2023 and 2024, so anything older than 2024 is meaningless.

Kenya is worth a separate note. Kenya’s National Research Fund runs no postdoctoral call at all. Neither does Nigeria. In Ghana, Ethiopia and Morocco I could not find a single published postdoc rate. In much of Africa the formal postdoc position barely exists — a new PhD goes straight to Lecturer. That absence is itself one of the findings of this issue, and it is not a small one.

The quiet good deals

If you are choosing where to apply, these are the places where the real value is much better than the reputation.

Poland. NCN grant-funded postdoc posts pay PLN 140,000 a year. That is $37,956 at market rates and $67,857 in real value — 11th out of 45, ahead of the Netherlands and the United States.

Taiwan. NTD 780,240 a year, which is only $24,526 on paper but $57,202 in real terms. Taiwan has one of the biggest gaps between how the number looks and what it does.

Hong Kong. The Research Grants Council fellowship pays HKD 420,090 plus a travel allowance. Eighth on real value.

China. A top-tier package at Peking or Tsinghua runs CNY 200,000 to 300,000, with housing on top. The national 博新programme pays CNY 280,000 a year for two years, plus a one-off CNY 80,000 research grant. Be careful: provincial universities pay CNY 120,000 to 200,000, so the spread inside China is about three times.

Malaysia, Thailand and Vietnam. All three are far better than they look. Two Thai schemes are worth knowing about: Mahidol’s international postdoctoral fellowship pays THB 60,000 a month, about 1.5 times the domestic rate, and it is open only to foreigners.

Brazil, but only in São Paulo. CNPq’s national scholarship pays BRL 5,200 a month. FAPESP, the São Paulo state funder, pays BRL 12,570 a month plus a 15% research allowance. That is 2.4 times more for the same career stage in the same country.

Eight questions to ask before you accept

Print this. Ask every one of them before you sign anything.

1. Is this a full position, or a percentage of one? Ask for the percentage in writing. In Germany especially.

2. Is this a salary or a stipend? A salary makes you an employee with rights. A stipend often does not.

3. What is deducted? Ask for an estimate of the monthly amount that will actually reach your bank account, not the gross.

4. Is there an employer pension, and what percentage? In Australia and Denmark this is worth around 17% of your salary.

5. Is housing included, provided, or my problem? Then find the actual rent for a one-bedroom flat near the campus and subtract it.

6. What is the visa tied to? If the visa ends with the contract, what is the notice period?

7. Does the offer include a flight home, health insurance, and relocation costs? Standard in the Gulf. Rare in Europe.

8. Is there any research money attached to me personally? Chile’s Fondecyt adds CLP 4.5 million a year. Belgium’s FWO adds €4,000 to €10,000. South Africa’s NRF adds ZAR 50,000. India’s ANRF adds ₹200,000. This money buys you independence, and people forget to ask about it.

How to read these numbers

I would rather tell you the limits than have you quote a number that does not hold up.

These are gross figures, before tax, except where marked tax-free. I did not model income tax in 45 countries. Tax rates depend on your marital status, your children, your residence, and treaties between countries. Table 5 tells you where tax matters most.

“Typical” means a defensible standard rate, not the only rate. I used national pay scales and official funder rates where they exist. Where a country has no national scale, I used the most widely applicable published rate and said so.

Real value uses the World Bank purchasing power parity factor for household consumption, 2025. I also checked every country against the Numbeo cost-of-living index. The two methods agreed within 20% for most countries. They disagree most on Turkey, Taiwan, Thailand, Singapore, Hong Kong, Vietnam, Malaysia and Egypt, so treat those ranks as approximate.

Some figures are weaker than others. The ones I would defend hardest are the published national scales: Germany, the Netherlands, Norway, Denmark, Ireland, Austria, Poland, Taiwan, Hong Kong, India, Israel, Brazil, Mexico, Chile, South Africa and the United States. The ones I would treat as approximate are Egypt, Pakistan, the Philippines, Vietnam and the Gulf states, where no official postdoc pay table is published. Nigeria and Kenya are Lecturer salaries used as proxies, because no postdoc scheme exists.

Exchange rates move. These are the rates on 22 August 2026. A single day’s rate can move a country several places, especially Turkey, Japan and Argentina. Argentina is not in the table at all, because its purchasing power figures have not been published since 2021 and any number I gave you would be wrong within a month.

Sources

In short, salary figures come from national pay scales and funder rate tables: the UK national pay spine 2025-26; TV-L E13 (Germany); CAO Nederlandse Universiteiten (Netherlands); SNSF and EPFL (Switzerland); FWO (Belgium); FWF (Austria); the IUA researcher scale (Ireland); Juan de la Cierva (Spain); FCT (Portugal); the 2025 post-doc decree (Italy); the state salary scale code 1352 (Norway); the AC agreement (Denmark); the YPJ system (Finland); NCN (Poland); Masaryk University (Czech Republic); TÜBİTAK 2218 (Turkey); NIH NOT-OD-26-044 (United States); the Canada Postdoctoral Research Award; Go8 academic scales (Australia); the University of Auckland academic agreement (New Zealand); Tsinghua and Peking University recruitment notices and the national 博新 programme (China); JSPS and RIKEN (Japan); IBS and the Sejong Science Fellowship (South Korea); NTU and NSTC-based university scales (Taiwan); the RGC Postdoctoral Fellowship Scheme (Hong Kong); NUS and NTU (Singapore); USM (Malaysia); Mahidol and Chulalongkorn C2F (Thailand); VINIF (Vietnam); ANRF, DST and CSIR (India); HEC (Pakistan); the American University of Sharjah, Khalifa University and MBZUAI (UAE); KAUST and KFUPM (Saudi Arabia); reported Doha salaries and HBKU (Qatar); Weizmann and Tel Aviv University (Israel); the NRF (South Africa); CONUASS (Nigeria); the UASU agreement (Kenya); CNPq and FAPESP (Brazil); SECIHTI (Mexico); ANID Fondecyt (Chile).

Cost of living: World Bank indicator PA.NUS.PRVT.PP, 2025 values, database updated 13 July 2026, cross-checked against the Numbeo Cost of Living Index, 2026 mid-year release. Exchange rates: 22 August 2026.

Before you go

If this was useful, forward it to the person in your lab who is about to accept an offer without asking question number one.

And if you are still working on your first paper, I wrote a book about that part. It is called How to Publish Your First Research Paper: The Unwritten Rules of Getting Accepted in International Journals, and it is on Amazon for $6.99. It is the thirty years of advice I usually give in my office, written down.

Next week I am looking at how long peer review actually takes, journal by journal. If you have ever waited eight months for a first decision, that one is for you.

Good luck with your applications.

— John

Short answer: research funding in the United States, the United Kingdom and Europe has tightened sharply since 2023. Over the same period China, Japan, South Korea and Taiwan expanded theirs, and a dozen faster-growing Asian countries did too. Many of their programs are open to foreign scientists. A few are reserved for them.

This is the full map. Who is spending. Who is growing fastest. What you would be paid. What that money buys where you would be living. And the twelve named programs a North American or European researcher can apply to right now.

The numbers in one place:

● The NIH R01, the main US research grant, funded 21.6 percent of applications in FY2023 and 13.0 percent in FY2025.

● In 2024, China spent more on research than the United States for the first time: about $1.028 trillion against $1.009 trillion in purchasing-power dollars.

● Japan’s main national grant, KAKENHI, funded 28.4 percent of applications in FY2025 — more than double the NIH rate.

● South Korea spends 5.13 percent of GDP on research, second in the world behind Israel.

● Indonesia published 1,392 indexed articles in 2010 and 38,992 in 2023, a 28-fold rise.

Picture two emails arriving in the same week. One is a grant rejection. The other is a job offer from a university in Asia, with the research money already approved. For a lot of researchers right now, that is not a hypothetical.

One thing I want to say up front. Most coverage of “research in Asia” means China, Singapore and the Gulf. Those are the rich ones. They are not the whole story, and in some ways they are not even the best part of it.

New here? I write one plain-English issue like this most weeks, for researchers working out how to publish, get funded and build a career across borders. It is free. Subscribe below and you will get the next one.

How bad are the research funding cuts in the US, UK and Europe in 2026?

Start with the United States. The R01 is the main NIH research grant. In FY2023, the success rate was 21.6 percent. In FY2025, it was 13.0 percent.

Look at how that happened. The mechanism matters. Applications went up 12 percent. Awards went down 22 percent. The squeeze is in the number of grants, not obviously in the dollars. NIH funded fewer, larger awards.

The National Science Foundation tells the same story another way. NSF made about 11,000 new research grants in FY2024. In FY2025 it made about 8,800. By April 2026, NSF had made only 613 awards. That is about a fifth of its normal pace for that point in the year.

Europe is not the easy option people assume. The ERC Starting Grant is the EU’s flagship early-career award. In the 2025 call, 3,928 researchers applied and 478 won. That is 12.2 percent.

The UK has been shedding people. Times Higher Education went through the published accounts of 90 universities. It counted roughly 29,550 staff taking severance across three years. The cost was about £591 million. More than a third of it came in a single year.

And it shows up in what researchers say about their own plans. STAT and the MassINC Polling Group surveyed 989 NIH-funded researchers in early 2026. 81 percent of junior tenure-track scientists said the upheaval put their tenure at risk. 53 percent had told their own students to look at jobs outside the US. 13 percent had already lost people to labs abroad.

The table in words. The NIH R01 success rate fell from 21.6 percent in FY2023 to 13.0 percent in FY2025. NSF new research grants fell from about 11,000 in FY2024 to about 8,800 in FY2025, and stood at only 613 awards by April 2026. The ERC Starting Grant funded 478 of 3,928 applicants in the 2025 call, or 12.2 percent. UK universities recorded about 29,550 severances over three years at a cost of roughly £591 million.

Which country spends the most on research and development?

In 2024, China spent more on research and development than the United States. About $1.028 trillion against $1.009 trillion. That is the first time in modern history. One thing to be precise about: these are purchasing-power dollars, which adjust for what money actually buys in each country. At market exchange rates the US still spends more.

Rewind, and the trend is even clearer. In 2000, China was 4.8 percent of world R&D spending and the US was 38.9 percent. In 2024, China was 29.6 percent and the US was 29.0 percent. The EU-27 fell from 23.6 percent to 17.6 percent over the same stretch.

Paper counts moved with the money. China now writes about 31 percent of the world’s science and engineering papers. The US writes 12 percent. India is third at 7 percent.

The Nature Index shows it in a single list. It ranks institutions by output in about 145 selected natural-science and health journals. So it is a narrower measure than QS or THE. In its 2026 tables, nine of the world’s top ten research institutions are Chinese.Ten years ago, in the 2016 tables, only two of the top ten were in Asia. Three were American.

The table in words. China’s share of global R&D spending rose from 4.8 percent in 2000 to 29.6 percent in 2024. The United States fell from 38.9 percent to 29.0 percent. The EU-27 fell from 23.6 percent to 17.6 percent. In absolute purchasing-power dollars for 2024, China spent about $1.028 trillion, the United States about $1.009 trillion and Japan about $234 billion.

Which Asian countries fund foreign researchers?

China. The NSFC raised its Excellent Young Scientists Fund by 50 percent. That is 12,000 more funded projects, starting in 2026. In late 2025 China launched a “K visa” for young foreign scientists. It covers AI, robotics and advanced materials. What makes it unusual is that you can relocate without a job offer in hand. And people are moving. The South China Morning Post tracked ten senior scientists who left US and UK universities for Chinese ones. That was in the first months of 2026 alone.

Japan. Japan spent ¥23.79 trillion on R&D in FY2024, about $149 billion at market exchange rates. (Table 2 above uses purchasing-power dollars, which put Japan at $234 billion. Same yen, measured differently.) That was up 7.9 percent. It is the highest total on record, and the fourth rise in a row. It works out to 3.70 percent of GDP.

Japan is also the clearest case of a country pointing money directly at you. Look at the JSPS Postdoctoral Fellowship, short-term. It is restricted to citizens or permanent residents of the US, Canada, the EU, the UK, Switzerland, Norway and Russia. You also qualify if you have done research continuously in one of those places for three years or more. Western researchers are literally the eligible class. It pays ¥362,000 a month, plus airfare, insurance and a settling-in allowance. There are three rounds a year.

And Japan’s odds are better. KAKENHI, the main national grant, funded 28.4 percent of applications in FY2025. That is more than double NIH’s 13 percent. Fair comparison caveat: KAKENHI covers many small grant categories, and the average award is far smaller than an R01. The odds are still not close. In July 2025 Japan also announced J-RISE. It is a ¥100 billion package built to hire from abroad. Separately, under Japan’s ¥10 trillion University Fund, Tohoku University was the first campus named a University for International Research Excellence. It received ¥15.4 billion for FY2026 and expects about ¥10 billion a year for 25 years. Its stated target is 500 international hires.

South Korea. Korea spends 5.13 percent of GDP on research. That is second in the world, behind only Israel. It is well ahead of every Western country. The 2026 government R&D budget rose 19.3 percent year on year.

The route in is the NRF Brain Pool program. It is aimed at scientists living outside Korea, of any nationality. It provides up to ₩300 million a year per researcher, for one to three years. That covers salary, relocation, children’s education and research costs. Korea also expanded its Top-Tier Visa in May 2026. If you qualify, you and your family get residency on arrival, and permanent residency in three years instead of five. Be realistic about the bar, though. It wants a major prize, Highly Cited Researcher status, triadic patents, or five years as a PI at a top-100 institution. Plus a Science Ministry recommendation. Korea plans to issue only about 350 of these by 2030.

Taiwan. Taiwan crossed NT$1 trillion in R&D spending for the first time in 2024. That was up 11.9 percent, and it works out to 4.1 percent of GDP. Only two countries spend a higher share. The Yushan Young Fellow program adds up to NT$1.5 million a year on top of the standard professor salary. There is a matching allowance for research and travel. The term is five years, renewable once, and it is open to international candidates.

Hong Kong. Hong Kong is small. It spends 1.13 percent of GDP. Its pitch is not size. It is odds. In June 2026, the Research Grants Council funded 206 of 617 Early Career Scheme applications. That is 33 percent. Its General Research Fund ran at 30 percent. Both held steady from the previous year. The Junior Research Fellow Scheme pays about HK$425,000 a year, for two to three years. There are 60 awards annually and no nationality requirement. But a local university has to nominate you. Find the host first.

Singapore. Singapore put S$37 billion into research through 2030. That is 32 percent more than its last plan. NTU’s Nanyang Assistant Professor scheme gives new hires a start-up grant of up to S$1 million. The teaching load is reduced too. Worth knowing, though. Singapore’s publication output has been flat since 2018. It is buying quality, not volume.

The Gulf. I teach at the American University of Sharjah, so I will tell you what I actually see. Researchers arrive here and get real research funding approved fast. Sometimes in weeks. They are not writing a dozen competing proposals first. KAUST in Saudi Arabia and MBZUAI in Abu Dhabi work the same way. The research money is attached to the position. You do not have to win a national grant lottery before you can start.

Where research is growing fastest: India, Vietnam, Thailand, Malaysia, Indonesia and the Philippines

This is the section I would read first, and it is the one that gets the least coverage anywhere else.

Look at how fast research output is growing outside the wealthy countries. Indonesia published 1,392 indexed articles in 2010. In 2023, it published 38,992. That is 28 times more in thirteen years. Indonesia now out-publishes Singapore roughly three to one.

Vietnam went from 1,096 articles to 10,645 over the same period — nearly ten times. The Philippines grew seven times.

India. India’s R&D spending reached about ₹2.45 trillion in FY24, roughly $25.6 billion, or 0.84 percent of GDP. In cash terms that is up 85 percent from ₹1.33 trillion four years earlier, before inflation. As a share of GDP it moved from 0.66 percent to 0.84 percent. One honest caveat. UNESCO’s comparable series still shows a lower figure for India, near 0.66 percent for 2020. Treat the exact percentage as contested. The direction is not.

India is third in the world for published research on Scopus, with 367,497 documents in 2024. It is third for highly cited papers, with about 19,000 in the global top 10 percent. Its national missions are large and specific on paper. ₹6,004 crore for the National Quantum Mission over eight years. ₹10,372 crore for IndiaAI over five. ₹10,000 crore for Biopharma SHAKTI, announced in February 2026. Now the caveat that matters. Only about ₹700 crore of the quantum money and ₹400 crore of the AI money had actually been released by early 2026. India’s new national research foundation spent nothing at all in its first two budget years. Announced is not the same as disbursed.

For a foreign researcher, the practical routes are short and well paid rather than permanent. GIAN brings overseas faculty in to teach short courses. It has approved 2,163 of them. It pays $8,000 for a 12-14 hour course, or $12,000 for 20-28 hours, travel included. SPARC funds research visits of one to four months. It pays $15,000 for the first month and $10,000 a month after. VAJRA is built for exactly this and pays the same rates. But its official page has not been updated in years. Confirm with DST that it is still running before you plan around it. Full faculty posts at the IITs and IISc are open to foreign nationals too. IIT Hyderabad’s October 2025 advert spells out the terms. Contracts of up to five years, no tenure, and security clearance from the Ministries of External Affairs and Home Affairs. There is no national rule here. Terms are set institution by institution, so read the actual advert.

Vietnam. Vietnam currently spends 0.41 percent of GDP on research. In December 2024 the Politburo wrote a target of 2 percent by 2030 into Resolution 57 — roughly a fivefold increase in six years. Decree 249/2025 has been in force since September 2025. It is the instrument for hiring foreign science talent. It covers a relocation grant, a housing allowance, and an annual round-trip flight for you and your family. It gives you a residence card matched to your contract, and an eased path to citizenship. Salary is negotiated, with no floor set. Vietnam’s stated priority is semiconductors. The government wants over 50,000 engineers and its first fab by 2030.

Thailand. Mahidol University runs an International Postdoctoral Fellowship that is restricted to non-Thai nationals. ฿60,000 a month, one year renewable to three, four intake rounds a year, with airfare, insurance and visa costs covered. Pair it with Thailand’s Long-Term Resident visa. Staff of Thai state universities and research institutes skip the income requirement entirely. They get a ten-year permit at a flat 17 percent income tax. Thailand’s real scientific strength is tropical medicine. MORU, the Mahidol-Oxford unit, is headquartered in Bangkok. Its network runs about 780 staff across five research units and 50 sites in Asia and Africa. It recruits internationally.

Malaysia. Malaysia publishes an actual international pay scale, which is rare and useful. Universiti Teknologi Malaysia’s DS51 international contract grade runs RM 5,855-12,445 a month in basic pay, plus housing, cost-of-living and service allowances. English is the working language. The national semiconductor strategy put RM 25 billion of fiscal support behind the sector and has drawn RM 70.7 billion in investment. Be aware the counter-trend: Malaysia’s R&D intensity has actually fallen, from 1.42 percent of GDP in 2016 to about 1.01 percent in 2022.

Indonesia. BRIN, the national research agency, runs a postdoctoral programme whose eligibility explicitly covers foreign nationals. You need an existing collaboration with a BRIN researcher. It runs one year, renewable once. The stipend amount is not published, so ask before committing. Indonesia’s LPDP education endowment stands at Rp 154 trillion, about $8.6 billion, with a ring-fenced research corpus of about $728 million. The scientific draw is tropical biodiversity, marine science, and nickel-based battery research.

The Philippines. I will be straight with you here. The flagship Balik Scientist Program is closed to you by statute — the law requires Filipino citizenship or Filipino descent. And the academic pay scale is genuinely low. An entry-level assistant professor earns about ₱42,178 a month in base pay, before statutory allowances and bonuses. Even so, base pay alone sits below the cost of living in Manila. The realistic route is IRRI, the International Rice Research Institute at Los Baños. It recruits globally on international contracts. It also holds the world’s largest rice genebank, with over 127,000 accessions.

Kazakhstan. A curiosity worth knowing about, rather than a recommendation. Kazakhstan has 758 researchers per million people, about three times India’s rate, though it spends only 0.162 percent of GDP on research. Nazarbayev University in Astana is English-medium, founded in 2010. Its published expatriate package includes rent-free housing and a two-course teaching load. Also health insurance, two flights home a year, relocation for dependents, and 56 days of paid home leave. Salaries are not published, so you will have to ask.

The table in words. Between 2010 and 2023, Indonesia’s indexed research output grew from 1,392 articles to 38,992, a 28-fold rise. Vietnam grew from 1,096 to 10,645, close to ten-fold. The Philippines grew about seven-fold. India published 367,497 documents on Scopus in 2024, third in the world.

What does a researcher get paid in Asia, and what does it cost to live there?

A salary figure on its own tells you almost nothing. What matters is the gap. What you earn, against what it costs to live there.

A foreign PhD hired at Ho Chi Minh City University of Technology earns about ₫54 million a month, roughly $2,062. A single person’s full monthly costs in Hanoi, rent included, run about $870. On Hanoi prices that is a buffer of nearly two and a half times. Ho Chi Minh City runs somewhat higher.

For comparison, the same $2,062 covers about 42 percent of a month in Boston.

India works the same way in reverse. To match the lifestyle of someone spending ₹180,000 a month in Bengaluru, you would need about $10,600 a month in San Francisco. This is why the short Indian programmes are interesting. GIAN, SPARC and VAJRA pay close to Western rates. You live at Indian costs.

The honest counterweight: local purchasing power in San Francisco is still higher in absolute terms. These are trades, not free lunches.

The table in words. A foreign PhD at Ho Chi Minh City University of Technology earns about ₫54 million a month, roughly $2,062, while a single person’s full monthly living costs in Hanoi run about $870 — a buffer of about 2.4 times. The same $2,062 covers roughly 42 percent of one month in Boston. Matching a ₹180,000-a-month lifestyle in Bengaluru would take about $10,600 a month in San Francisco.

12 research programs in Asia open to foreign scientists

The table in words. Japan’s JSPS Postdoctoral Fellowship (short-term) pays ¥362,000 a month plus airfare and insurance, and is open only to citizens or permanent residents of the US, Canada, the EU, the UK, Switzerland, Norway and Russia. South Korea’s NRF Brain Pool provides up to ₩300 million a year for one to three years, any nationality. Taiwan’s Yushan Young Fellow adds up to NT$1.5 million a year on top of a standard professor salary for five years. Hong Kong’s Junior Research Fellow Scheme pays about HK$425,000 a year for two to three years, 60 awards annually, no nationality requirement, university nomination needed. Thailand’s Mahidol International Postdoctoral Fellowship pays ฿60,000 a month and is restricted to non-Thai nationals. Singapore’s Nanyang Assistant Professorship carries a start-up grant of up to S$1 million. India’s GIAN pays $8,000 for a 12-14 hour short course or $12,000 for 20-28 hours, and SPARC pays $15,000 for a first month of research visit and $10,000 a month after.

Two things I would flag about this table.

First, notice how many of these are open to any nationality, or aimed at Westerners specifically. The assumption that these systems are closed to outsiders is mostly wrong. The JSPS short-term fellowship and the Mahidol postdoc are only open to foreigners.

Second, notice the deadlines. Several land soon. JSPS’s Standard fellowship — a longer scheme than the short-term one in the table — closes 28 August 2026 for its first FY2027 call. Taiwan’s Yushan second round runs in early September. Hong Kong’s Junior Research Fellow Scheme closes 30 October 2026. The JSPS short-term fellowship in the table has no single deadline; it runs three rolling rounds a year.

Does working abroad make you a better researcher?

This is the argument I care about most, and it does not appear on any funding table.

Researchers who live and work somewhere genuinely different from home tend to think more creatively. People who live somewhere genuinely different score higher on creativity tasks. That is a measured finding in the psychology research, not a travel-brochure feeling. And the researchers took care to show it is not simply that adaptable people are the ones who move. Adapting to a place that works differently seems to change how you approach a problem.

Asia makes this point better than anywhere else. Asia is not one thing. A lab in Tokyo is nothing like a lab in Bangalore. Or a lab in Hanoi. Or a lab in Abu Dhabi. Different funding structures. Different questions people think are worth asking. Different assumptions about how a research group should work.

So the benefit is probably not a single jolt. Each of those systems would stretch a different part of your thinking. That last step is my own inference, not something the studies measured. But the countries that get written about least are the ones furthest from what you already know. That is not a coincidence.

What to check before you move: language, contracts, family and exchange rates

Every research system has constraints. You do not escape them by moving — you swap one set for another. The things that decide whether a move works are usually dull ones:

● Language. English is the working language of research in Singapore, Hong Kong, Malaysia, the Gulf, and most Indian institutes. It is not the daily language in Japan, Korea, Vietnam, Thailand or Indonesia. The science is in English. The department meeting often is not.

● Contract length. Many of these are fixed-term by design. India caps foreign faculty contracts at five years, with no tenure. Most fellowships run one to three years. Know whether you are taking a stint or a career move.

● Coming back. Ask early how the work will count if you return. Grants, ethics approvals and lab kit rarely travel with you. Co-authorships do.

● Family. This is usually what decides it, not the salary. Check school costs, spousal work rights and visas for dependents before anything else. Korea’s Top-Tier Visa and Vietnam’s Decree 249 both cover families. Others do not.

● The exchange-rate trap. The yen and the won are historically weak right now. Converting those salaries into dollars makes them look worse than they actually live. Always compare a salary to local rent, not to your current salary.

Common questions

Which Asian country has the best odds of winning a research grant?

On published success rates, Hong Kong and Japan. Hong Kong’s Research Grants Council funded 33 percent of Early Career Scheme applications in June 2026 and 30 percent of its General Research Fund. Japan’s KAKENHI funded 28.4 percent in FY2025. Both are roughly double the NIH R01 rate of 13.0 percent.

Are research jobs in Asia open to foreigners?

Often yes, and sometimes only to foreigners. Japan’s JSPS short-term postdoctoral fellowship is restricted to citizens and permanent residents of Western countries. Thailand’s Mahidol International Postdoctoral Fellowship is restricted to non-Thai nationals. South Korea’s Brain Pool, Taiwan’s Yushan Young Fellow and Hong Kong’s Junior Research Fellow Scheme all accept any nationality.

Which country spends the most on research and development?

Measured in purchasing-power dollars, China, at about $1.028 trillion in 2024 against the United States at about $1.009 trillion. At market exchange rates the United States still spends more.

Do I need to speak the local language?

Not for the science. English is the working language of research in Singapore, Hong Kong, Malaysia, the Gulf and most Indian institutes. In Japan, Korea, Vietnam, Thailand and Indonesia the research is in English but daily department life often is not.

Is the pay worse than in the US or Europe?

In dollar terms usually yes, in living terms often no. A foreign PhD in Ho Chi Minh City earns about $2,062 a month against Hanoi living costs of about $870. The same $2,062 would cover about 42 percent of a month in Boston.

What is the catch?

Contract length, mostly. India caps foreign faculty contracts at five years with no tenure, and most fellowships run one to three years. Announced money is also not always disbursed money — India released only about ₹700 crore of its ₹6,004 crore quantum mission by early 2026.

What to do this week

Do one small thing. Pick one country from this list and find one program in your own field. Not to apply. Just to see what is there. Most researchers have never looked. That is part of why the odds are better in some of these places.

Wherever you end up working, publishing well is part of the plan. That is exactly what my book covers. It’s called How to Publish Your First Research Paper. The full title adds The Unwritten Rules of Getting Accepted in International Journals.It’s written in plain English, from years spent as a reviewer and an editor. It’s on Amazon now for $6.99.

If you know a colleague quietly weighing a move like this, forward this to them.

Good luck out there.

John

Picture two researchers. Both just finished a strong study. Same topic. Same quality of work. Same top journal in their field.

One works in Boston. The other works in Lagos.

They both hit submit on the same day.

Whose paper is more likely to get in?

If you guessed it’s not an equal fight, you’re right. And the size of the gap might surprise you.

The odds are not equal

Look at medicine first. Researchers studied 10,558 papers from five of the biggest medical journals in the world. Journals like The Lancet and The New England Journal of Medicine. The papers were published over ten years. The United States and the United Kingdom alone wrote 64 out of every 100 papers. Only 32 countries published more than 10 papers each over that entire decade. Nearly 200 countries exist in the world.

Now look at economics. In the ten most cited economics journals, the United States alone wrote 65 out of every 100 papers. Every developing country in the world, combined, wrote just 7. And this is not because developing countries do little research. They produce close to 40% of the world’s economy. But only 7% of the papers in the top journals.

One more number. Out of every 1,000 authors published in the journal Nature, only 8 came from the poorest countries on Earth.

These are not small gaps. And here is the part that really matters. The gap does not shrink as you move toward better journals. It grows. Researchers tracked this over 40 years. They found real progress for developing countries. But almost all of it happened in lower-ranked journals. The very top journals barely moved at all.

Why does this keep happening? Look at who decides

A paper does not publish itself. Someone reads it. Someone decides if it is good enough. So we looked at those people too. The editors who sit on the boards of these same top journals.

The pattern repeats.

At The Lancet and The BMJ, about 9 out of every 10 editors are based in the US or the UK.

At the top ten economics journals, most editors are American too. We checked all ten of those journals for editors from Africa, South Asia, and almost all of Latin America. We found one. At a university in Brazil.

To be fair, it is not the same story everywhere. One of those ten journals, the Review of Economic Studies, is run mostly by editors in Europe, not the US. So this is not one simple, tidy pattern. But the overall picture is clear. The people deciding what counts as good research come from a very small part of the world.

This helps explain why the gap does not close on its own. The researchers writing the papers and the editors judging them often come from the same small group of countries.

Authors vs. editors, side by side

Where you still have a real shot

None of this means you should give up on top journals. It means you should be smart about where you aim.

Some publishers already know this system is unfair. They built in real fixes. Here are a few worth knowing.

Most journals in the Nature family will publish your paper for free if you are the lead author from a low-income or lower-middle-income country. Not every Nature journal does this, so check first. You do not even need to apply. It happens automatically, for about 70 countries.

PLOS, another major publisher, does the same for the poorest countries, in any of its journals.

eLife will consider a full fee waiver for anyone, from any country, no matter your reason. You just have to ask.

But be careful here too. Most of these programs only cover the poorest countries. They use strict income cutoffs. A country like Brazil often falls into a gap. Too rich to qualify for the waiver. Too expensive to pay the fee without one. If you’re unsure whether your country qualifies, check the exact rules on the journal’s own website. Don’t assume.

Which publishers actually waive fees

The takeaway

The system is not fair yet. The data is clear about that. But it is not evenly closed either. A few real doors are open, and most researchers never even look for them.

Know the odds. Know which journals already made room for someone like you. And send your best work there.

If you’re also thinking about grant money, not just journals, my earlier newsletter on what actually gets a grant funded covers similar ground. Know the real rules of the game before you play it.

Getting your paper accepted anywhere starts with the basics too. My book, How to Publish Your First Research Paper, covers that groundwork in plain English. It’s on Amazon for $6.99.

If you found this useful, forward it. Send it to one colleague who is deciding where to submit their next paper.

Good luck out there.

John

Should I leave University?

I have been asked this question for thirty years. It used to come at the end of office hours, when the student had already put their bag on their shoulder and was standing by the door. That is usually when the real question arrives.

For a long time I gave an honest but useless answer. I said it depends. So I decided to go and find the actual numbers, from the countries where you are reading this. Some of them I expected. One of them I did not.

Here is what I found.

The cleanest comparison in the world

Most pay comparisons are unfair. They compare a professor in one country with an engineer in another. The two jobs are not the same job.

The United States government solves this problem for us. It publishes the pay for a single job title, then splits that same title by who the employer is. Same work. Same country. Same year.

The result is hard to argue with.

(Figure 1 — same job title, two employers. Median pay in the United States, May 2024.)

A medical scientist at a university earns a middle salary of $67,280. The same title at a commercial research company earns $121,240.

In computer science the gap is wider. $85,290 at a state university. $237,990 at a software company. Almost three times more, for the same job title.

That is the number people share. It is real. But it is American.

Europe: a smaller gap than you think

Europe does not look like the United States. Public pay scales are published, so nobody has to guess, and the distance between the two worlds is much shorter.

(Table 1 — academic pay and industry pay in Europe. Before tax, in local money.)

In Germany a postdoc on the public scale earns about €57,000 a year before tax. A PhD holder in the chemical industry has an agreed floor of €86,075. That floor is a real number, not an estimate. Employers and the union agree it, and it holds for 2025 and 2026.

So the German gap is about one and a half times. Not three.

In the United Kingdom it is smaller again. A postdoc sits between £37,694 and £50,253. The national middle pay for a biochemist or biomedical scientist in the whole economy is £47,892. At that level the two are almost the same. The gap only opens later, when industry offers a director role and the university offers another fixed-term contract.

Poland is the country that stopped me. From January 2026, the lowest a public university may pay an assistant professor is 7,044.50 złoty a month. The national minimum wage that same year is 4,806 złoty. So a person with a doctorate, teaching at a university, has a legal pay floor about 46% above the floor for any job at all.

Asia: the widest gaps on this list

(Table 2 — academic pay and industry pay in Asia. Before tax, in local money.)

China gives us the sharpest single fact in this whole newsletter.

A study published in 2025 put the average pay for faculty at Chinese research universities at ¥155,400 a year. In the same period, a recruitment report looked at new PhD graduates entering companies. They started at an average of ¥25,082 a month. That is about ¥300,000 a year.

Read those two numbers again. A fresh graduate in a company starts at roughly double what the average professor earns.

Korea shows the other side of the same coin. One third of new PhD holders there were not employed at all in the most recent survey. Not underpaid. Not employed. The government’s answer was a new postdoc programme paying ₩90 million a year. That is more than double the average at the country’s science institutes. Countries do not build programmes like that unless people are leaving.

India is the case I get asked about most. An assistant professor starts at a basic salary of ₹57,700 a month, and allowances are added on top of that. A data scientist in the private sector averages about ₹1.2 million a year. The gap is real, but it is not the four or five times that people repeat to each other.

Where the university pays more

Now the part I did not expect.

(Table 3 — three countries where academic pay is the higher number.)

In Brazil, a new federal professor with a doctorate, on a full-time exclusive contract, earns R$13,753.96 a month. The government publishes the table, so there is no argument about it.

The average pharmacist working in drug research and development in Brazil earns R$7,773.96 a month. That figure comes from the government’s own employment records, not a survey.

The university pays nearly twice as much.

Colombia is similar. A professor at a public university can reach 8 to 11 million pesos a month, with a doctorate and a research record. Private universities generally pay less, not more.

South Africa is a third case. A lecturer at Wits starts at a package of R761,000 a year. The average monthly earnings in the whole formal economy are R29,490, which is about R354,000 a year. An entry-level academic earns roughly twice the national average.

I want to be careful here. This does not mean Brazilian or South African academics are comfortable. It means the sentence “industry pays more” is not a law of nature. It is a description of one country. And there is a real chance it is not a description of yours.

Africa: the gap is not industry, it is the exit

In several African countries the comparison that matters is not university against company. It is staying against leaving.

(Table 4 — pay, strikes, and staff losses.)

The University of Lagos lost 1,800 of its 3,000 academic staff over five years. The 1,200 who stayed now teach more than 50,000 students. The union estimates that more than 5,000 lecturers leave the country every year. More than 2,300 Nigerian academics have moved to the United Kingdom since 2021. Reported offers from abroad are five to ten times local pay.

In Kenya, lecturers were on strike for 49 days across 42 public universities, ending in November 2025. The dispute was about money owed under an agreement from 2017.

In Ghana, one lecturer with a doctorate said publicly that he earned less as a university teacher at home than he had earned as a graduate student in the United States. He left.

These are not pay comparisons any more. They are a country losing the people it trained.

What the direction of travel says

One last number, and it is the one I would keep.

In 2004, 56% of new American PhD holders who had a job lined up went into academia. By 2024 that was 40%. Industry now takes exactly the same share. In computer science, 61% went to industry last year alone.

Nobody announced this. It happened one graduate at a time, over twenty years.

Five questions before you decide

1. Compare the same job title, not the same country. Ask what your exact role pays at a company in your city. National averages hide almost everything.

2. Compare the whole package. Housing, pension, health cover, tax, and how many months of pay per year. In Switzerland and the Netherlands the extra months are a large part of the total. In Nigeria and India, allowances can be worth more than the basic salary.

3. Ask about year two, not year one. Industry often negotiates the first year freely and then applies an agreed floor from the second. Germany’s chemical industry does exactly this.

4. Price the contract, not just the salary. A permanent post at a lower number can beat a higher number on a two-year contract, especially if you need a visa, a mortgage, or a stable place for your family.

5. Check whether your country is Brazil. Before you accept that industry pays more, look up your own national tables. In several countries the university is the higher-paying option, and nobody is telling you.

One more thing

If this was useful, forward it to one person who is deciding right now. That is how this newsletter grows, and it stays free either way.

And whichever side of this you choose, published papers are the currency in both. That is what my book is about. It is called How to Publish Your First Research Paper: The Unwritten Rules of Getting Accepted in International Journals. It is on Amazon, and it costs less than a coffee in most of the countries above.

The honest summary is this. There is no worldwide answer to “should I leave”. There is only your country’s table, and most people never look at it.

Now you have.

See you at the next one,

John

This week’s video went through all three stages of a research career: PhD student, postdoc, and professor. It showed the shape of the gap. This newsletter has the actual numbers, country by country, for all three.

Every figure below comes from an official source: a funder’s own page, a government pay scale, or a university HR table. I checked them all this month (August 2026). Exchange rates move, so treat the USD column as a guide, not an exact number. Where a figure is an advertised range and not an official scale, I say so.

One more thing before the tables. I have been paid to do science in Canada and in the Gulf. I have also supervised students who were paid one tenth of what my students in North America were paid, for the same work. None of this is about merit. It is about geography and policy. Knowing the numbers is how you stop being surprised by them.

The Video:

What PhD students earn

Let’s start at the bottom of the ladder. That is where the differences are biggest, and where most of you reading this are standing.

There are three situations, and they are very different.

1. You are an employee. You get a salary, a pension, and a raise every year. The raise is written into a national agreement, so nobody has to grant it.

2. You get a stipend or fellowship. It is usually smaller. Sometimes it is tax-free. It only goes up when a government decides to raise it, which can take ten years.

3. There is no national funding at all. You pay for yourself, or you hold a job and do the PhD around it. This is normal in much of the world.

A few notes that don’t fit in a table:

The United States is not what people assume. Stipends run from about $20,000 to $50,000, depending on the university and the field. The NIH minimum is $29,364 (FY2026). The well-known NSF fellowship pays $37,000. But two PhD students checked 178 American biology programs, and in only 2% could a student cover rent and food in that city. In physics, three-quarters of departments pay first-year teaching assistants less than that, even if they share an apartment. A famous name on the building does not mean the student can afford to live there.

Germany’s job ads need translating. German PhD researchers are paid on the public pay scale, called TV-L E13. But the contract is usually part-time: 50% in the humanities, 65–67% in the natural sciences, 75–100% in engineering and computer science. A 65% contract is about €3,000 per month before tax. So when you see “E13 65%” in an advert, that is what it means.

In China, the extra money is the real money. The national grant for doctoral students is ¥13,000–15,000 per year. Yes, per year. What students actually live on is the extra money added by the university and the supervisor. That usually brings it to around ¥3,000 a month, and top universities pay several times more. So ask the specific lab, not about the country.

India and Brazil show the cost of waiting. India’s JRF has been ₹37,000 a month since March 2023, and the next increase has still not arrived. Brazil’s doctoral scholarship did not rise once between 2013 and 2023, a full ten years, before a 40% increase. Prices do not stop rising while a committee decides.

Nigeria, and much of Sub-Saharan Africa, has no national PhD stipend at all. The main public funder, TETFund, pays for university staff to do PhDs, and has funded more than 23,000 lecturers since 2011. But a typical doctoral student pays for themselves, or works a job at the same time. If that is your situation, you are doing something harder than most of the people who give you career advice ever did.

What postdocs earn

The gap doesn’t close after the PhD. It follows you.

The one to know about is the Marie Skłodowska-Curie fellowship, or MSCA. It pays a living allowance of €5,990 a month, plus €710 a month to help you move. The living allowance is then adjusted for the country you go to. Note that this is the amount paid to the university, so employer taxes come out before your salary.

It is one of the best-paid early-career positions in the world. And you can apply from any country, including developing countries. If you remember one thing from this issue, remember to check whether you can apply for an MSCA.

One more: Switzerland’s national funder sets postdoc salaries at CHF 80,000–110,000. That is not a typing error. It is also why a Swiss postdoc advert gets 300 applications.

What professors earn

Three things the averages hide:

The American number is really two numbers. The average full professor at a US research university earns about $181,000. But about two-thirds of US academic staff no longer have a permanent position. They are paid for one course at a time, at an average of about $4,000 per course. Teach eight courses in a year and you have earned $32,000, with no benefits and no security. “US professor salary” describes two very different jobs with one name.

In Germany, the listed salary is only the starting point. The W2 and W3 salaries in the table are the base. On top of that comes performance pay, which you negotiate yourself, and which adds several hundred to over a thousand euros a month. In Germany a professorship is negotiated, so go in prepared.

Nigeria’s professors lost two-thirds of their pay without a pay cut. In 2023 the naira lost most of its value against the dollar. In dollar terms, a professor’s salary fell from about $1,300–1,800 a month to about $400–550. The number printed on the payslip never changed. A 40% raise arrived in January 2026, and it did not come close to repairing the loss. Currency decisions are salary decisions.

A smaller number can be more money: the package math

This is the section that changes decisions, so let’s do the arithmetic properly.

KAUST in Saudi Arabia pays PhD students $25,000–30,000. In the table above that looks average. Now add everything else the offer includes.

Free housing on campus. Health insurance for you and your family. No tuition fee. No income tax. Moving costs paid. KAUST says the full package is worth $70,000–80,000 a year. Once you subtract Zurich rent from a Swiss salary, that average-looking stipend can leave you with more money in your pocket.

Faculty offers in the Gulf work the same way. A tax-free salary, plus housing (often 40–100% of the rent), school fees for your children, flights home once a year, and a payment when you leave the job. Those benefits add 30–50% to the package. I live inside one of these packages, so I will say it directly: if you compare a Gulf offer to a European one using base salary only, you are comparing the wrong numbers.

The rule: compare what is left after rent, tax and school fees, not the number at the top of the offer.

And the second rule: a dollar is not a dollar. Exchange rates do not tell you what daily life costs. An entry-level professor in India earns what converts to about $37,000, but prices there are lower, so it buys roughly three times more. A Swiss salary looks enormous until you look at rent in Zurich. Economists call this correction “purchasing power parity.” You can just call it “check the rent first.”

What actually moves researcher pay

Here is the pattern behind all the numbers above. Pay changes when policy changes. And policy changes when people push.

Graduate workers at the University of California held the largest academic strike in US history, and raised minimum pay by nearly 50%. The UK raised its stipend by the largest real amount since 2003, after the cost of living became a national story. Canada nearly doubled its doctoral award in 2024, after ten years of campaigning. France put the doctoral minimum into law, so it now rises automatically. Brazil’s ten-year freeze ended because of public pressure, not generosity.

None of these were rewards for good science. They were responses to organised pressure. Remember that the next time someone tells you researcher pay is set by a market.

Five questions to ask before you accept any offer

1. What is the official rate, and where is it published? Ask for the pay-scale page or the funder’s notice, not the number in the advert. If nobody can show you one, that tells you something too.

2. Is it a salary or a stipend? A salary usually includes pension payments, sick leave, parental leave and a yearly raise. A stipend often includes none of these. A tax-free stipend can still beat a taxed salary, so check the tax and the benefits together.

3. What comes with the money? Housing or a housing allowance, health insurance (does it cover your family?), tuition, moving costs, flights, school fees. Ask for each one in writing.

4. What will be left after rent? Find the real rent in that city, not the average for the country. One number can change the whole decision.

5. When was this rate last increased, and how does it increase? A stipend frozen since 2023 in a country with rising prices is a falling salary with the same name on it. Automatic yearly steps, as in the Netherlands, Germany and France, are worth real money over four years.

Your three-line worksheet

This is the assignment from the video, with a little more detail. For any position you are considering, write down:

1. The official pay, taken from the official source

2. The local rent for a small flat, or a shared flat, in that exact city

3. The average salary in that country, so you know how you would live compared to the people around you

Line 1 minus line 2 tells you whether you can live there. Line 3 tells you how it will feel. A university ranking tells you neither.

Bookmark these

These are the official pages I used. They are worth saving:

• UKRI stipend (UK): ukri.org → “Get a studentship”

• NIH NRSA stipends (US): the annual NOT-OD notice (FY2026 is NOT-OD-26-044)

• NSF GRFP (US): nsf.gov GRFP solicitation

• MSCA rates (EU): the Horizon Europe MSCA work programme / unit-cost decision

• SNSF salary ranges (Switzerland): snf.ch, “salary ranges” annex

• EPFL doctoral salaries: epfl.ch HR pages

• Dutch university CAO scales: universiteitenvannederland.nl

• TV-L tables (Germany): oeffentlicher-dienst.info

• French doctoral contract minimum: legifrance.gouv.fr (arrêté of 29 Aug 2016, as amended)

• Canada CGS-D: cihr-irsc.gc.ca tri-agency pages

• CSIR/UGC fellowship rates (India): csirhrdg.res.in

• KAUST fellowship: admissions.kaust.edu.sa → Fees & Funding

If this issue was useful, send it to one researcher who is deciding about an offer right now. That is exactly who it is for.

One more thing. Everything above is about where you are standing on the ladder. Moving up it usually comes down to publications. If your first international paper is the thing in the way, I wrote a book about exactly that: How to Publish Your First Research Paper: The Unwritten Rules of Getting Accepted in International Journals. It is on Amazon now — https://www.amazon.com/dp/B0H4RM4NXD — and I wrote it for researchers who are doing this without much help, and in a second or third language.

See you next week.

John

This week’s video made one argument: grant funding is partly a lottery, but three things measurably move your odds. This newsletter is the part I couldn’t fit in a short video: the actual numbers, funder by funder, checked against their own published data, plus the templates I promised. If you’re writing an application this year, keep this one

.

The odds, honestly

Let’s start with the numbers most funding workshops skip.

In the 2025 fiscal year, the NIH funded 13.0% of R01-equivalent applications. One year earlier it was 18.7%. That is a collapse of almost a third, in twelve months, driven by budget turmoil and a surge in applications. Europe is not much friendlier: the 2025 ERC Starting Grant call funded 12.2% of proposals, the Consolidator call 11.2%, and the latest Marie Skłodowska-Curie Postdoctoral Fellowship round funded 9.6% of over 17,000 applications. The friendliest of the big systems, the NSF, funded 29% in its last fully reported year, and UKRI sat at 22%.

And if you’re applying from Africa through a program like ARISE, the pilot call drew close to 1,000 applications for 44 grants. Roughly 4%.

Two things follow from this table. First, single-digit and low-teen success rates mean that most rejections say nothing about your quality: the NSF itself reported that in one year alone, researchers requested $3.8 billion for more than 3,900 proposals that were rated as high as or higher than the average funded proposal, and were declined anyway. Money ran out before merit did. Second, if rejection is the statistically normal outcome even for excellent proposals, then your strategy cannot be “write one perfect grant.” It has to be “maximize odds per attempt, and plan for more than one attempt.” Everything below serves that.

What actually moves your score

First, the principle this channel keeps coming back to: science is about the question. A grant is an argument that a gap in our knowledge matters and that you can close it. No method, however elegant, rescues a proposal that never names its gap. Hold onto that while you look at the strangest finding in the grant literature, because the two fit together better than they first appear.

Researchers at the NIH analyzed 123,707 R01-equivalent applications and asked which of the five review criteria — significance, investigator, innovation, approach, environment — actually determines the overall score and the funding outcome. The result was lopsided. A one-point improvement in the approach score raised the odds of funding by a factor of 6.2. The next strongest criterion, significance, managed 2.1. Everything else trailed behind.

Does that mean the big idea doesn’t matter? No. It means the approach section is where reviewers test whether your question is real. A vague question can’t generate a sharp plan, and a sharp plan is the most convincing evidence the gap exists and can be closed. The question earns the reading; the plan earns the money. Practically: the gap gets named on page one, and the methods section gets a disproportionate share of your writing hours, because that’s where belief is won or lost.

What does a six-times approach section contain? Five things reviewers consistently reward:

• Aims that visibly close the named gap. Each aim should read as a step toward answering the question, not a showcase for a technique you happen to own. If an aim would survive with the gap deleted, it’s a method looking for a job.

• A design someone could execute from the page. Not “we will investigate X” but which experiments, in which order, with which samples, analyzed how. If a knowledgeable stranger couldn’t sketch your Gantt chart from your text, it’s too vague.

• Feasibility evidence. Preliminary data if you have it; if you don’t, cite your own related methods papers, name the facilities and collaborators, and show the arithmetic that your budget and timeline can actually deliver the plan.

• A pitfalls-and-alternatives paragraph for every aim. This is the paragraph most first-time applicants leave out, and its absence is one of the most common weaknesses reviewers flag. “If the field experiment fails to establish, we will use the greenhouse system already running in our lab” costs you three lines and buys you credibility.

• Scope discipline. Overambition is a feasibility problem wearing an ambition costume. Three aims that obviously fit the budget beat five aims that might.

The person you’re writing for

Here is the uncomfortable science on reviewers. When 43 experienced reviewers were given the same 25 funded NIH applications, their agreement on scores was statistically zero. Not low. Zero. They didn’t even agree when translating the same listed weaknesses into numbers. And a separate analysis of 102,740 funded NIH grants found that the percentile score a grant received explained about 1% of the variation in what that grant went on to produce. The system is noisy, and the people running it know it.

Reviewers are also stretched thin. In the largest global survey of grant review, reviewers reported spending about ten days per year on all reviewing combined, across every proposal assigned to them.

You cannot fix any of that. What you can do is write for the reviewer you’ll actually get: intelligent, rushed, reading adjacent to their specialty, and required to summarize you to a room. Practically, that means:

• Your summary/aims page carries the whole application. Many panelists form their score there and read the rest to confirm it.

• Headings, topic sentences, and one figure that explains the design. A reviewer skimming at 11 p.m. should collect your three main claims from the bolded text alone.

• Give your champion their lines. In panel, one advocate speaks for you. Write the three sentences you’d want them to say, and make sure those sentences appear, nearly verbatim, on page one.

• Confidence, calibrated. An analysis of tens of thousands of funded and unfunded proposals found more promotional, confident language associated with meaningfully higher funding odds. It’s correlation, not magic, and hype without substance backfires. But chronic hedging reads as doubt, and non-native English writers hedge most of all. “We will determine” beats “we hope to explore.”

Here is a skeleton for that page-one summary. Steal it:

• Paragraph 1 — the problem: two sentences on the gap, one on why it matters beyond your subfield, and to whom.

• Paragraph 2 — the project: “We will [verb] X by [approach], which will establish Y.” Name your aims in one sentence each.

• Paragraph 3 — why us, why here: the one dataset, method, site, or collaboration that makes your team the right one. Local advantage counts double: access nobody else has is a moat, not a consolation prize.

• Paragraph 4 — the payoff: what exists at the end that doesn’t exist now, and who uses it.

The resubmission effect

Now the numbers I most wish someone had shown me thirty years ago.

A study followed 11,808 first-time NIH applicants whose proposals were rejected. Among those whose application was at least discussed by the panel, 81% resubmitted. Among those triaged without discussion, only 41% ever came back. The ones who came back were rewarded: resubmitting the same project, rather than starting fresh, multiplied the odds of funding within three years by up to 4.2, and more than half of the discussed-and-rejected group eventually got that very grant funded. NIH’s own data book tells the same story year after year: resubmissions succeed at roughly double the rate of first submissions.

Think about what that 41% number means. The single most common response to a triaged application is to abandon it, which is precisely the move the data says is wrong. Rejection isn’t the end of the pipeline. Statistically, it is the pipeline.

The resubmission playbook, condensed:

1. Wait two weeks before deciding anything. Reviews read as insults on day one and as data on day fifteen.

2. Sort every comment into three bins: fixable (methods detail, missing citation, unclear aim), arguable (a judgment call you can defend), fatal (eligibility, fit, scooped). Most applications have zero fatal comments.

3. Fix everything in bin one, even the ones you disagree with. If one reviewer misread it, the next one will too. Misreading is information about your writing.

4. Answer bin two respectfully in the response document — never sarcastically, and never by ignoring it.

5. Resubmit to the same scheme if rules allow; if not, the revised application goes to the next-best funder on your list. The work is never wasted.

What a proposal really costs (and what that means for strategy)

Some honesty about the price of playing. Researchers self-report an average of 116 principal-investigator hours per proposal, and about 171 hours once co-investigators are counted — roughly a month of full-time work. An Australian study measured 38 working days per new proposal and calculated that a single national funding round consumed about 550 working years of researcher time, for a 21% success rate. And in a large US survey, funded investigators reported spending 44% of their research time on administration rather than research.

One more finding from that 116-hours study, and it’s a strange comfort: the amount of time spent on a given proposal had no measurable relationship with whether it was funded. Past a competent threshold, more polishing hours don’t buy more odds. What does scale your odds is attempts and fit. The strategic conclusions:

• Recycle relentlessly. Aims pages, methods text, figures, budgets: a rejected proposal should supply 60-70% of the next one. Your second application should cost half the hours of your first.

• Match effort to expected value. A 10% mega-grant and a 40% small national grant deserve very different shares of your month.

• Run a portfolio. Several small and medium streams beat one giant application in both odds and survivability — the argument of the funding-cut video, and it applies before the crisis, not just after it.

The part that isn’t your fault

The video’s reality check, with its receipts. Within Europe’s funding system, researchers based in the “widening” countries — mostly Central and Eastern Europe — win ERC grants at 6.8%, against 15.0% for the older member states. Same competition, same forms, half the odds. And a major survey of scientists across eight countries found non-native English speakers were 2.5 times more likely to have work rejected over the writing itself, and spent about half again as much time writing in English — a tax paid before a reviewer ever judges the science. That study measured papers rather than grants, but nobody who has sat on an international panel believes the effect stops at journals.

I include this not to discourage you but because pretending the field is level is its own kind of lie, and because the practical response follows directly: apply where your odds are structurally better (national schemes, South-South programs like TWAS, seed grants with regional eligibility — the full funder map is in the Video 6 newsletter), use every legitimate language tool to write in confident, plain English, and let the portfolio, not any single verdict, carry your career.

The email almost nobody sends

Most funders’ program officers will tell you, on the record, that they wish more applicants contacted them before applying — and that remarkably few do, especially from outside North America and Western Europe, where many applicants assume it’s not allowed. It is allowed. It is encouraged. Ten minutes of a program officer’s honest reaction can save you a month of misdirected work, because fit — applying to a scheme that was never going to fund your topic — kills quietly and early.

Copy, adapt, send:

Subject: Fit inquiry — [two-word topic] for [program name]

Dear Dr. [name],

I’m a [career stage] at [institution], working on [one plain-English sentence: what and why it matters].

I’m considering applying to [program] in the [month] round, with a project that would [one sentence: core aim and approach]. Before I invest the full application, may I ask whether this falls within the program’s current priorities — and if not, whether another program in your portfolio would be a better fit?

I’d be glad to send a one-page summary if useful. Thank you for your time.

[Name, position, institution, one relevant link]

Send it four to six months before the deadline. If the answer is “not really a fit,” you just saved a hundred hours. That’s the highest hourly rate you’ll ever earn.

The pre-submission checklist

Print this. Ten minutes, the night before the internal deadline, not the real one.

1. Could a non-specialist state my aims after reading page one only?

2. Does every aim have a method, a timeline, and a pitfalls-and-alternatives plan?

3. Would a stranger believe my team can do this — and does page one say why?

4. Is there one figure that explains the whole design?

5. Have I cut hedging language from the summary (”hope to,” “attempt to,” “might”)?

6. Do budget, timeline, and aims obviously match?

7. Have I checked eligibility and formatting against the current call, not last year’s?

8. Has one person outside my field read the summary page and passed the three-question test?

9. If this is a resubmission: is every reviewer comment either fixed or answered?

10. Do I know exactly which funder this goes to next if it’s rejected?

That last question is not pessimism. As the numbers above show, it’s the single best predictor of eventually getting funded.

See you next week.

All figures above come from the funders’ own published statistics and the peer-reviewed studies named in the text, checked in early August 2026.

Hi,

Last issue was about the rules: what to disclose and where, so your paper doesn’t get desk-rejected. I promised the harder conversation would come next, and here it is. Not “how do I avoid getting caught,” but “when is leaning on AI actually honest, and when does the work quietly stop being mine?”

The new video makes the case in a few minutes:

Watch it first. Then keep this newsletter. It holds what the video couldn’t: the full survey numbers, the reader-trust experiment, the three tests as a card you can print, and eight real situations with a straight verdict on each.

What we all say, and what we all do

In 2025, Nature surveyed about five thousand researchers across disciplines and countries on one question: is it OK for AI to write a science paper? The answers, side by side, tell you exactly where our community’s conscience currently sits.

Look at the shape of it. Almost everyone is comfortable with AI polishing or translating writing. Two thirds accept AI helping to draft, with disclosure. Opinion turns negative when AI writes a referee report. And then the last row: only about 8% say they have actually used AI to help write a paper, and many of those never told anyone.

That last row is the whole story. We have collectively decided the tools are mostly acceptable, and we are still hiding them. There’s a word for doing something you believe is fine while making sure nobody finds out, and that word is not “fine.” It means our instincts haven’t caught up with our policies. The three tests below are my attempt to close that gap: they’re the questions I ask myself, and the ones I give the researchers I mentor.

The three tests

Every AI decision in research fits one of three questions. If a use passes all three, use the tool with a clear conscience and a one-line disclosure. If it fails any one, don’t do it, or change how you’re doing it until it passes.

Test 1: The substitution test

If someone took the tool away, would the thinking still exist?

This is the test that separates expressing your ideas from outsourcing them. Grammar, phrasing, translation, structure suggestions on an argument you built: take the tool away and your science is intact, just rougher in English. That passes. A hypothesis you can’t explain the origin of, a discussion section whose argument only exists because a chatbot proposed it, a literature synthesis of papers you never opened: take the tool away and there’s nothing left. That fails, no matter what you disclose.

Notice that the Nature respondents drew this same line without being told to. Comfort is near-total at the language end, splits in the middle at drafting, and turns to resistance as AI approaches the ideas themselves. The community’s collective gut already knows where the line is. The test just makes it usable at the moment you’re typing a prompt.

One practical corollary: AI is at its most legitimate when it’s working on something you have already produced. Give it your rough draft, your bullet points, your reasoning to tighten. The more your prompt looks like “here is my thinking, help me say it,” the safer you are. The more it looks like “give me thinking,” the deeper you’re wading.

Test 2: The out-loud test

Could you describe exactly what the tool did, in plain words, to your co-authors and in the paper itself?

“I used DeepL to translate my own draft from Bahasa Indonesia.” Easy sentence. “ChatGPT wrote my introduction and I lightly edited it.” Harder sentence. “The related-work section is a chatbot summary of papers I haven’t read.” Nobody will say that sentence out loud, and the refusal is diagnostic. If describing a use embarrasses you, the problem is the use, not the disclosure.

This test has a second job: it produces your disclosure statement for free. The sentence you’d comfortably say out loud is, almost word for word, the sentence that belongs in your acknowledgments or methods (last issue has the copy-paste template and every publisher’s required location). Write it before you use the tool, keep it in the file, and you’ll never have to reconstruct what you did months later.

Test 3: The accountability test

Can you personally stand behind every sentence, citation, and figure, with your name and your career?

Every major publisher, without exception, refuses to list an AI tool as an author, and they all give the same reason: authorship means accountability, and a model cannot be accountable for anything. That refusal tells you how editors think about everything else. The moment you submit, there is no “the AI did it.” Every reference it suggested, every number it summarized, every confident claim it made is now yours, personally.

So the operating rule is simple: nothing goes in the file that you haven’t verified yourself. Not “it looked right.” Verified. This is the test that catches the failure modes from last issue — invented references, fabricated details, altered images — before they catch you.

What readers actually think when you disclose

Here’s the study I found most useful this year, because it answers the question researchers actually worry about: “if I disclose, will people trust my work less?”

A preregistered experiment published in AI and Ethics took 806 members of the public, demographically representative of the US, and described the same piece of research to them six different ways: no AI at all, AI language polishing, AI polishing plus an explicit note that all the intellectual work was human, AI helping with theory, AI helping with methods and analysis, and AI helping with both.

Three findings worth your attention.

First, yes, trust was highest when no AI was involved, and it fell as AI moved deeper into the intellectual work. Readers hold roughly the same instincts researchers do.

Second, and this is the practical one: when the disclosure said AI polished the language and made clear that the thinking, analysis, and conclusions were entirely human, ratings recovered most of the way back toward the no-AI level. The damage isn’t done by the tool. It’s done by ambiguity about who did the thinking.

Third, people who used AI often themselves showed much smaller trust penalties. As these tools normalize, the cost of honest disclosure is already shrinking, and it will keep shrinking. The cost of being caught hiding something is not shrinking. It never does.

The lesson for your next paper is concrete: don’t just write “the authors used ChatGPT.” Write what it did and what it didn’t do. “We used [tool] to improve the readability of the manuscript. All study design, analysis, interpretation, and conclusions are the authors’ own.” That second sentence is doing more work for your credibility than anything else on the page.

Eight grey zones, eight straight verdicts

These are real situations, most of them from emails and comments I’ve received. The table gives the verdict; the notes below give the reasoning for the contested ones.

“AI drafted my introduction from my detailed bullet points.” This one divides people, so here’s my honest take: it can pass, on three conditions. The bullet points contained the actual ideas and structure. You reworked the output until it argues the way you argue. And you verified every factual claim and citation it touched. If your “bullets” were two vague lines and the machine invented the framing, it fails Test 1, because the thinking wasn’t yours.

“AI suggested my hypothesis.” As a brainstorming partner throwing possibilities at you, fine — researchers have always used conversations to spark ideas. The line is what happens next. If you took the suggestion, grounded it in literature you read, connected it to your data, and can now defend it as your own reasoned position without mentioning the chatbot, it has become your hypothesis. If you can’t explain why it should be true beyond “the AI said it was plausible,” you’re presenting machine output as your intellectual contribution. That fails both Test 1 and Test 3.

“AI summarized the papers in my literature review.” Using AI to find and triage papers: good, that’s search. Citing papers you know only through an AI summary: fails Test 3, full stop. Models misread papers and invent findings, and you’re accountable for every characterization. Read what you cite. No exceptions, including the exception you’re thinking of right now.

“AI wrote my referee report.” The majority of researchers in the Nature survey disapproved, and the publishers are harsher: pasting someone’s unpublished manuscript into a chatbot is a confidentiality breach nearly everywhere. Beyond the rules, notice what was asked of you: your expert judgment. Outsourcing it fails Test 1 by definition. Using AI to polish the wording of the judgments you formed yourself, without uploading the manuscript, is where the defensible line sits.

“AI wrote the science case in my grant.” Funders are now the strictest players in the system. NIH stopped treating substantially AI-developed applications as the applicant’s original work in September 2025, and other funders are following. Reviewers are funding your mind. On the science case, the substitution test applies with no mercy: budget tables and formatting, fine; specific aims, no.

The pattern across all eight: the verdict is never about the tool. It’s about which part of the work the tool touched, and whether a human mind you can name did the thinking and checked the claims.

The fairness question

I want to end where I always end, with the researchers this channel is actually for.

These tools are the most levelling technology to arrive in academic writing in my thirty years. If you think in Urdu or Amharic or Vietnamese and publish in English, AI assistance can finally let your ideas compete on their merits instead of their grammar. The Nature survey found non-native English speakers are already among the heaviest users, and that’s rational, and it’s right.

But the same tools are flooding journals with hollow papers. At least one in eight biomedical abstracts in 2024 already showed signs of AI writing. Analyses of recent retractions attribute the largest share to compromised peer review and paper mills, and the flood is making editors more suspicious, not less. Here’s the unfair part, and you already know it: when trust erodes, suspicion doesn’t land on famous labs with famous names. It lands on unknown researchers from institutions editors don’t recognize, in countries they’ve never visited. The people with the least margin absorb the most doubt. We saw the same asymmetry with AI detectors last issue, falsely flagging 61% of non-native writers.

Which means every honest disclosure you write does more than protect your paper. It’s evidence, one paper at a time, that researchers from your institution, your country, your community use these tools the right way. That reputation is a commons. The three tests are how you avoid draining it and how we slowly refill it.

One honest word

I ran this newsletter through the tests myself, since you should know how I work. I use AI tools weekly: to check my reasoning, to tighten drafts I’ve written, to find sources I then read. The arguments here are mine, every study was verified against the original source, and a human being you can name stands behind every claim. That’s not a boast. That’s just what the contract with a reader looks like now, and it’s a contract worth keeping.

If you haven’t watched the video yet, please go ahead and do so. And if you’re not subscribed, subscribe below so the next issue lands in your inbox.

Next issue is the one I’ve been wanting to write for months: AI as the great equalizer. If English isn’t your first language, you’ve spent your whole career being judged on your grammar instead of your science. That’s finally changeable. Which tools actually help, how to use them without tripping any line in this issue, and how to keep your own voice while you do it. The hopeful one, next time.

Good luck out there, and see you in the next issue.

— John

The Research Roadmap is free. If a colleague or student would find this useful, forward it — that’s how it grows.

Sources

• Is it OK for AI to write science papers? Nature survey shows researchers are split — Nature news feature, 2025 (~5,000 researchers; >90% accept AI editing/translation; ~65% accept AI-assisted drafting; majority disapprove of AI referee reports; ~8% have used AI to write papers, many undisclosed) — https://www.nature.com/articles/d41586-025-01463-8

• When researchers use AI: public trust, ethical judgments, and the perceived value of academic research — AI and Ethics, 2026 (preregistered experiment, n=806) — https://link.springer.com/article/10.1007/s43681-026-01039-w

• Delving into LLM-assisted writing in biomedical publications through excess vocabulary — Kobak et al., Science Advances, 2025 (≥13.5% of 2024 abstracts) — https://www.science.org/doi/10.1126/sciadv.adt3813

• GPT detectors are biased against non-native English writers — Liang et al., Patterns, 2023 / Stanford HAI (61% false-flag rate) — https://hai.stanford.edu/news/ai-detectors-biased-against-non-native-english-writers

• Academic journals’ AI policies fail to curb the surge in AI-assisted academic writing — PNAS, 2025 (~70% policy adoption; ~0.1% disclosure) — https://www.pnas.org/doi/10.1073/pnas.2526734123

• Learning from Retractions to Drive Prevention Strategies — MDPI/Crossref retraction data analysis, 2025 (compromised peer review + paper mills = largest share of retraction reasons) — https://blog.mdpi.com/2026/07/23/retractions-data-2025/

• ICMJE — AI use by authors (accountability/authorship) — https://www.icmje.org/recommendations/browse/artificial-intelligence/ai-use-by-authors.html

• NIH — NOT-OD-25-132 (AI-developed applications not treated as original; effective 25 Sep 2025) — https://grants.nih.gov/grants/guide/notice-files/NOT-OD-25-132.html

Hi,

If you’re using AI anywhere near your research writing, and by now most of us are, there’s one mistake that can get your paper desk-rejected before a single reviewer reads it. It isn’t using AI. It’s hiding it, and pointing it at the wrong part of your paper.

I just released a video on exactly this — and it pairs with everything below.

Watch the video first (it’s short). In a few minutes it walks you through the one rule that keeps you safe.

Then keep this newsletter. This is the complete, checked version: every major publisher’s real rule in one place, the exact sentence you can paste into your manuscript, the stricter rules for grant applications, and the trap that wrongly flags non-native English writers. It goes well beyond what the video could hold, and it’s built to come back to every time you sit down to submit.

Everything below was checked against the publisher’s own page while I was writing it. Let me start with the number that made me want to write it.

The gap:

First: roughly 70% of journals now have an AI policy. Second: in a study of 75,000 recent papers, only about 0.1% actually disclosed using an AI tool. That’s 76 papers. Out of seventy-five thousand.

Read those top two rows together. The rule exists almost everywhere. Almost nobody is following it. Most of them aren’t cheating. They just don’t know the rule is there, so they walk straight into a problem they never saw.

The people this hurts most are exactly the people I made this channel for: researchers writing in their second, third, or fourth language, often without a senior mentor to say “wait, did you disclose that?” So here is the mentor version, in writing.

Rule 1: Disclose it — one sentence, correct place

Editors are not trying to catch you for using a language tool. They’re trying to see that you were honest about it. The disclosure is not a confession. It’s the thing that protects you.

Two questions decide where the sentence goes:

1. Did AI only help with language (grammar, phrasing, readability)? → It goes in your acknowledgments (or a short AI-declaration section).

2. Did AI touch your data, your analysis, or your figures? → It goes in your methods, described specifically.

At most journals, if you only used AI for basic grammar and spelling, you don’t have to declare anything at all. The declaration kicks in the moment AI helped generate text, ideas, structure, or anything in the science itself.

The publisher-by-publisher table

This is the reference. Find your target journal’s publisher and read across.

A few things jump out when you line them up. Nature wants the disclosure in one place. Science wants it in three, and treats getting it wrong as the same category of offense as faking data. Grammar-only help is usually exempt. And every single one of them says the same thing about authorship: an AI tool can never be listed as an author, because authorship means being accountable, and a chatbot can’t be accountable for anything.

The sentence you can copy

You don’t have to invent the wording. Elsevier publishes a template, and it’s clean enough to adapt for almost any journal. Keep it, change the brackets, done:

During the preparation of this work the author(s) used [NAME OF TOOL / SERVICE] in order to [REASON]. After using this tool/service, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the published article.

Worked examples you can lift:

• “…used ChatGPT (GPT-4) in order to improve the readability and grammar of the manuscript. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article.”

• “…used Grammarly in order to correct grammar and spelling.” (At most journals this one is exempt anyway — but a line costs you nothing and removes all doubt.)

• “…used DeepL Write in order to improve the English of the Discussion section, which was originally drafted by the authors in [language].”

Notice what the sentence does: it names the tool, names the narrow job, and states that a human checked it and owns it. That last clause is the whole point. It’s you telling the editor, “a person is responsible here.” Write it before you write the paper and keep it in the file from day one, so you never have to reconstruct what you did.

Rule 2: Never let AI touch what it can’t be responsible for

This is the rule that actually ends papers, so I want to be specific about the failure modes.

Fabricated references. This is the big one. Chatbots invent citations that look completely real: plausible authors, a plausible journal, a plausible year, a DOI that leads nowhere. A reviewer only has to check one, find that it doesn’t exist, and now every other claim in your paper is suspect. This isn’t hypothetical anymore. In 2026, a major research venue announced it “has been desk-rejecting, and will continue to desk-reject” papers containing hallucinated references or false claims — no warning, no chance to fix it. If AI drafted any part of your text, check every reference against the real source yourself before you submit.

Data and analysis. Don’t let a general chatbot run or interpret your statistics and then present the output as findings. If AI was part of your analysis pipeline, that’s a methods-section disclosure with enough detail that someone could reproduce it, not a quiet shortcut.

Images and figures. Nature and Science ban AI-generated images by default. Taylor & Francis forbids using AI to add, hide, move, or remove any feature in a figure. Elsevier bans AI-made graphical abstracts outright. A diagram, a flowchart, or a schematic is usually fine with disclosure. A “microscopy image” or a “result” that came out of an image generator is a fast route to a misconduct finding.

Here’s the clean line to hold in your head:

AI is allowed to help you say things better. It is not allowed to decide what’s true. That’s your job, and it’s the job you get accountability for.

Rule 3: The detector trap — and why it targets you specifically

Now the part that’s genuinely unfair, and the reason this topic matters more for our audience than almost anyone else.

Some editors and institutions run submissions through “AI detectors.” Those tools do not work reliably, and they fail in one specific, discriminatory direction. A Stanford study ran essays through seven popular detectors:

You can write every word of your paper yourself, in your own hand, and still be accused of using AI — purely because English isn’t your first language. The detectors read simpler sentence structure and a smaller vocabulary as “machine-like,” and that’s exactly what second-language writing often looks like.

You can’t stop an editor from running a detector. But you can make the accusation collapse the moment it’s raised. Three defenses:

1. Keep a paper trail. Draft in Google Docs, or Word with version history on, or a Git repo. If you’re ever challenged, you can show the document growing over days — messy middle drafts, tracked edits, your comments to yourself. No AI dump looks like that. This single habit is the strongest protection you have.

2. Keep your own voice. The harder you sand your writing down into generic, polished, over-formal English, the more it reads like a machine — and ironically, the more likely a detector is to flag it. Some roughness and personality is not a weakness here. It’s evidence you’re human.

3. Disclose honestly and let it work for you. A clear disclosure line (“I used X to improve grammar”) reframes the whole conversation. You’re not hiding anything, so there’s nothing to catch.

Grant applications are a different, stricter game

Everything above is about journal papers. Funders play by tougher rules, and this matters enormously for our audience, because a rejected paper you can resubmit — a burned grant relationship you often can’t.

The clearest example is the US NIH, which changed its policy for applications submitted from 25 September 2025 onward. Two things you need to know:

The reason NIH gave was blunt: they were worried about a flood of AI-generated proposals drowning the system. So the line for grants is stricter than for papers. AI can help you understand a call, tidy your English, or check your budget math. It should not be writing your case. The ideas, the significance, the innovation — reviewers are funding yourmind, and funders now say so explicitly.

And the confidentiality rule is even harder here than in journals: if you review for a funder, do not put an applicant’s proposal into any AI tool. Reviewers are handling unpublished, fundable ideas that aren’t theirs. NIH and NSF both treat that as a serious breach. When in doubt, treat every proposal you’re sent as if it were locked in a drawer.

If you’re applying to European, Gulf, or South-South funders (the ones I covered two issues ago), check each call’s specific AI clause — they’re being added fast and they’re not all the same. The safe default everywhere: your ideas are yours, AI polishes the wrapper, and you say so if asked.

Quick answers to the questions I get most

“If I only used Grammarly, do I have to disclose?” Usually no. Most publishers exempt basic grammar and spelling correction. But a single honest line costs nothing and ends the argument before it starts. When unsure, disclose.

“Will disclosing make reviewers think less of my work?” No. Reviewers judge the science. A clean disclosure reads as professionalism, not weakness. What damages you is being caught not disclosing — that reads as hiding something.

“I used AI to translate my own draft from my language into English. Is that allowed?” Yes, at essentially every journal, with a one-line disclosure. This is exactly the use case these tools are best at. Say the draft was authored by you in your language and translated/improved with the tool.

“Can I use AI to summarise papers for my literature review?” To find and orient, yes. But read every paper you cite yourself and verify every claim. Never cite a source you only know through an AI summary — that’s how fake and misread references get in.

“What if I’m falsely accused because English isn’t my first language?” Point to your paper trail (drafts, version history) and your disclosure. This is a real and documented bias, not you doing something wrong. Evidence that your document grew over time is the strongest answer there is.

“Is any of this actually enforced, or is it theatre?” Increasingly enforced. In 2026 a major venue began desk-rejecting papers with fabricated references outright, and NIH now routes AI-generated applications toward misconduct review. Treat the rules as real.

Your 5-minute checklist before you hit submit

1. Found the target journal’s AI policy in their author guidelines? (Every major publisher has one.)

2. Written your one-sentence disclosure, in the right section for how you used AI?

3. Checked every reference against the real source?

4. Confirmed no AI-generated data, results, or figures (or gotten explicit editor approval)?

5. Got a paper trail ( version history or drafts) you could show if challenged?

The ethics underneath the rules

I want to step back from the rulebook for a moment, because every policy above is really just downstream of one idea, and it’s worth naming plainly.

Publishing rests on a quiet promise: that the thinking is yours, that the results are real, and that a human being stands behind every claim and can answer for it. That’s why every publisher, without exception, refuses to list AI as an author. It isn’t a technicality. Authorship means accountability, and a model can’t be accountable for anything. The disclosure rules, the image bans, the reference checks all trace back to protecting that one promise.

So the honest way to think about it isn’t “what can I get away with,” it’s “am I keeping that promise.” Three lines are worth holding firm:

• Ideas. Use AI to express your thinking, not to replace it. Passing off machine-generated reasoning as your own intellectual contribution is the line NIH now polices for grants, and it’s the one that quietly hollows out a scientific record.

• Results. Never let AI generate, alter, or “clean up” data, images, or findings. This isn’t a grey area. Every publisher treats it as fabrication, the most serious charge in science.

• Others’ trust. If you review or edit, the manuscripts and proposals you handle aren’t yours to feed into a tool. Confidentiality is an ethical duty, not just a rule.

There’s a fairness dimension too, and it cuts both ways. Used well, these tools genuinely lower a barrier that has kept good science from people who happen to think in another language. Used badly, the same tools can flood journals with plausible-looking, low-quality, or fabricated work, and it’s honest researchers, especially early-career ones without a name to protect them, who pay the price when trust erodes. Which side of that you land on is a choice you make every time you open the tool.

One honest word

I use these tools too. When at least one in eight papers already shows signs of AI help, pretending it’s not happening is silly, and pretending it’s all bad is worse. Used the right way, this is one of the most levelling things to arrive in academic writing in my career, especially for people the system was never built to accommodate. The researchers who get burned aren’t the ones who use AI. They’re the ones who never learned where the line was.

If you’re not subscribed yet, subscribe below so the next one lands in your inbox.

Next issue goes straight at the harder question this one only touched: the ethics of AI in research. Not how to avoid getting caught, but where the real lines are. When leaning on AI is honest and when it quietly isn’t, what we owe the people who read and trust our work, how to think about authorship and originality when a machine helped, and how to use these tools in a way you’d be comfortable defending in public. It’s the conversation the rulebooks are dancing around.

— John

The Research Roadmap is free. If a colleague or student would find this useful, forward it.

Sources

• Academic journals’ AI policies fail to curb the surge in AI-assisted academic writing — PNAS, 2025 (5.2M papers / 5,000 journals; ~70% policy adoption; ~0.1% disclosure) — https://www.pnas.org/doi/10.1073/pnas.2526734123

• Delving into LLM-assisted writing in biomedical publications through excess vocabulary — Kobak et al., Science Advances, 2025 (≥13.5% of 2024 abstracts; up to 40%) — https://www.science.org/doi/10.1126/sciadv.adt3813

• GPT detectors are biased against non-native English writers — Liang et al., Patterns, 2023 / Stanford HAI (61% TOEFL false-flag rate) — https://hai.stanford.edu/news/ai-detectors-biased-against-non-native-english-writers

• ICMJE — AI use by authors — https://www.icmje.org/recommendations/browse/artificial-intelligence/ai-use-by-authors.html

• Elsevier — Generative AI policies for journals (incl. disclosure template) — https://www.elsevier.com/about/policies-and-standards/generative-ai-policies-for-journals

• Springer Nature / Nature — AI editorial policy — https://www.nature.com/nature-portfolio/editorial-policies/ai

• Science / AAAS — Editorial policies (generative AI) — https://www.science.org/content/page/science-journals-editorial-policies

• Taylor & Francis — AI policy — https://taylorandfrancis.com/our-policies/ai-policy/

• ICLR 2026 — Response to LLM-generated papers and reviews — https://blog.iclr.cc/2025/11/19/iclr-2026-response-to-llm-generated-papers-and-reviews/

• NIH — NOT-OD-25-132, “Supporting Fairness and Originality in NIH Research Applications” (6-application cap; AI-developed applications not treated as original; effective 25 Sep 2025) — https://grants.nih.gov/grants/guide/notice-files/NOT-OD-25-132.html

Hi — good to have you back.

The video this week makes one argument: the money didn’t leave science, it moved, and most of us are still looking north out of habit.

This is the part the video can’t do. Below is the actual list. Names, amounts, deadlines, and — the part nobody else will give you — which programmes are dead, so you don’t spend a Saturday writing into a portal that no longer exists.

Everything here was checked against the funders’ own pages in the week of 14 July 2026. Where I couldn’t confirm something, I say so instead of guessing. Funding pages change; treat this as a map, not a contract, and always open the official page before you write anything.

One warning before we start. This list is long on purpose. Do not read it all. Find the two or three rows that describe you, and ignore the rest.

First: the graveyard

Start here, because this is where most people waste their time. Every one of these still shows up on funding blogs, university newsletters, and “top 20 grants for African researchers” listicles. All of them are dead, paused, or gutted.

And one that isn’t dead but is wobbling: Fulbright. The FY2026 budget proposed a 93% cut and elimination; Congress restored the exchange bureau at $667M (down ~$74M), after a payment freeze and a dozen board members resigning in 2025. The Visiting Scholar and Foreign Student programmes technically still run. Expect smaller cohorts and delays. Verify before you invest weeks in it.

I’m not listing these to depress you. I’m listing them because the average “funding opportunities” post you’ll read this year is somewhere between two and five years out of date, and it costs you real time. If a list you find doesn’t mention that TWAS paused its flagship grants this year, that list wasn’t checked.

The one rule that explains almost everything below

Before the tables, learn this and you’ll read any funding page in about forty seconds.

Ask: who is allowed to be the applicant?

Not who benefits. Not who can be “involved.” Who signs. Every programme on earth falls into one of four buckets:

Most of the disappointment I see in early-career researchers comes from spending three weeks on a host-led programme while believing it was direct. The Gulf is where this happens most, so I’ve made that explicit below.

1. Direct: apply yourself, stay where you are

This is the shortest list in most people’s heads and it shouldn’t be.

On Wellcome, an important date. Right now, researchers at organisations across LMICs can hold Wellcome awards (mainland China and India excluded). From 29 October 2026, eligibility narrows to the UK plus LMICs in Africa, South Asia and South-East Asia. If you’re in Latin America, the Middle East, or Eastern Europe, your Wellcome window is closing this year. (Ireland gets cut too, which tells you this is a budget decision, not a Global South decision.) That’s not a rumour; it’s on their eligibility page. Act accordingly.

2. Direct-ish: you’re a real co-applicant, but you need a northern partner

Different from “host-led” — here you are a genuine grant-holder with your own budget. You just can’t start alone.

3. Europe’s consortium money — and the list nobody told you about

This is the single most valuable thing in this newsletter.

The rule: an entity from outside Europe can normally join a Horizon Europe project but cannot take EU money.

The exception: the European Commission publishes a list of low- and middle-income countries whose entities are automatically eligible for EU funding as full beneficiaries. The current version is v3.8, dated 12 February 2026, and it contains 117 countries. Kenya, Nigeria, Pakistan, Peru are all on it. Egypt came off it when it became an Associated Country — which tells you the list moves, so check the live PDF rather than trusting anyone’s memory, including mine.

Find it by searching: “List of Participating Countries in Horizon Europe” on the EU Funding & Tenders portal.

If your country is on that list, then every time a European colleague has told you “we’d love to include you but there’s no money for partners outside Europe,” they were wrong, or they hadn’t looked.

How you actually get into a consortium, since this is the question everyone asks:

CORDIS — search funded projects in your field, find who coordinated them, email the coordinator.

The EU Funding & Tenders Portal “Partner Search” — needs an EU Login, and people genuinely post there.

Your National Contact Point — most countries have one, including many in Africa, Asia and Latin America. They run brokerage events. Almost nobody uses them.

Be in the room before the call opens. By the time a call text is public, the team usually exists.

Two Horizon myths to kill:

MSCA Postdoctoral Fellowships are “open to any nationality” — true — but you must physically move to a host in Europe. €399.05M this year, ~1,600 fellowships, deadline 9 September 2026. It is not money for staying home. The “Global” variant is for Europeans going out, not for you coming in.

The Hop-on Facility is not for you. It’s for institutions in EU “widening” countries. Third countries are explicitly excluded. I’ve watched people try.

4. Mobility: go, get funded, come home

Finite stays. You keep your job. You come back with a skill, a network, and sometimes equipment.

5. The Gulf: read this before you write anything

I live and work in this region, so let me save you a month.

Every Gulf sovereign programme uses the same model: a local lead PI, and you as a capped international collaborator. The money is real and growing — Saudi Arabia’s R&D spending hit SAR 29.48 billion in 2024, up 30.4% in one year, the newest official figure — but it is buying capability for the country paying.

So the Gulf is not a grant you apply for. It’s a relationship you build, and then a grant. The realistic entry points, in order: the Grand Challenges competitions (open to anyone), a visiting professorship (needs a nominator), co-supervising a student, or being the named specialist on someone else’s project. And whatever you send, read the national strategy first — this money buys stated priorities, and a generic proposal loses to someone who did the reading.

6. Free things, if your budget is genuinely zero

The APC waiver is the one people miss. If you’re a corresponding author in a Group A country, you may be able to publish open access for nothing, at journals you assumed were out of reach. Check your country’s group before you next pay a fee, or before you next assume you can’t afford OA.

7. What’s actually ahead of you

Everything below is still in front of us as of mid-July 2026. Confirm each on the official page — dates move.

8. Where to start, depending on who you are

On SGCI and ARISE, since the video mentions both. Be precise about what they are:

The Science Granting Councils Initiative third phase runs 2026–2030, roughly US$42M (CAD 57M), backed by Canada, Germany, the UK, Norway, South Africa and Wellcome, across 20 sub-Saharan councils. But SGCI funds councils, not researchers — its current live call is for an institutional capacity hub (max CAD 500,000). The money reaches you through your own national science council’s calls. So the action item is: find out what your national council is running. Most researchers I meet cannot name it.

ARISE funded 47 PI-fellows across 38 African countries at up to €500,000 each. That’s the original cohort. I could not confirm an open call for new applicants as of this week; a second phase has been pledged. Watch list, not to-do list.

Now the part that matters more than the list

I’ve just given you seventy-odd programmes. I want to be honest about what that’s worth.

Almost nothing, on its own.

I’ve met researchers with beautiful funding spreadsheets. Forty programmes, colour-coded, deadlines, eligibility, links. And zero applications submitted. The spreadsheet had quietly become the work. It feels like progress and it costs nothing, which is exactly why it’s so appealing when you’re frightened about money.

Look back at the tables and you’ll notice something. The largest amounts are all in the “consortium” and “host-led” rows — and every one of those has a human being standing in the door. Horizon Europe doesn’t fund countries, it funds consortia, and consortia are assembled from a phone call between people who already know each other, months before the call text is public. The MSCA is not a fellowship you win; it’s a supervisor who agrees to spend a month writing with you. CAS PIFI says it outright: no direct applications, you need a host. ASPIRE’s visiting professorship needs someone in Abu Dhabi to nominate you. Even the ERC’s new African Union arrangement is a door that only opens if someone on the other side knows to open it.

The money moved. The mechanism didn’t. It still travels along relationships.

Which means the sequence most of us were taught is backwards. It isn’t “find the grant, then find the people.” It’s the reverse. The people come first. The grant is what you do together once they exist.

So: send one email this week. One. To one human connected to one of the rows above. Not a proposal, no CV attached. Two paragraphs — here’s who I am, here’s the specific thing I noticed about your work, here’s a question. That’s the whole assignment.

I know that email feels harder than three hours of spreadsheet. That’s the tell. The thing that feels hardest is almost always the thing that’s actually load-bearing.

And one last word about the graveyard at the top. Look at what killed most of those programmes: Newton, GCRF, FLAIR, the Commonwealth fellowships, the ASM award, Fulbright’s near-death, the pressure on TWAS’s Sida grants and OWSD’s fellowship. Nearly all of it traces back to two governments changing their minds. That’s the whole argument of this week’s video in one table. If your career runs on one country’s political weather, you are not doing research, you’re placing a bet.

The fix isn’t finding a better single funder. It’s building several small streams that no single signature can switch off. Rufford plus Idea Wild plus a society grant plus your national council is not a glamorous portfolio. It is a very hard one to kill.

Do me a favour: hit reply and tell me the one programme on this list you’d never heard of. Whichever one you all name, that’s the next deep dive. I read every reply.

Talk soon, John

Deadlines and eligibility change without much warning, and a few entries above are marked uncertain on purpose. Always open the official page before you write. If you find something here that’s out of date, tell me and I’ll correct it for everyone.

If someone forwarded this to you and you want it every week, you can subscribe here: https://researchroadmap.substack.com

Subscribe / read every week: https://researchroadmap.substack.com

Hi! Good to have you back!

I want to start with one number: 86%.

That’s the share of USAID’s awards that were terminated after early 2025 — roughly 5,300 grants and contracts, somewhere around $80 billion in cancelled funding, with another $9 billion taken back later in the year. About 94% of the agency’s staff were let go. And if you look at all of US foreign assistance together — around 11,000 contracts and grants across USAID and the State Department — only about 28% were still active by the middle of 2025. So close to 7 in 10 were gone.

I used a bit of this in this week’s video, but I kept the numbers light there, because nobody wants a lecture read off a spreadsheet. Here I want to give you the fuller picture, because I think the scale actually helps. When your own grant gets cut, it feels personal, like you did something wrong. The numbers say otherwise. This did not happen to you. It happened to the whole system at once.

A few more figures worth knowing:

• The NIH funded about 3,600 foreign sub-awards in a single recent year, worth more than $400 million — the exact channel that paid for a lot of labs in the Global South. That model has now ended.

• Around 15% of all NIH grants had an international component. This was never a side activity. It was one in seven.

• In a national survey, a quarter of NIH-funded scientists said the change hit them significantly.

• The areas hit hardest are the ones that mattered most to people’s lives — HIV, malaria, tuberculosis, maternal health, cancer. One analysis put the annual damage to global-health research alone at up to $74 million a year if the restructuring proves hard to work around.

• And these are real projects, not line items: a cholera-prevention trial in Bangladesh, malaria and nutrition programs in Nigeria, Malawi, Cameroon, and Somalia — stopped partway through, with people already enrolled.

I lay all this out because the size of it is the first thing that should change how you feel. A cut this broad is not a signal about the quality of your science. It’s a policy decision made thousands of kilometres away, and it swept up strong labs and weak ones without looking at either. Once you really take that in, you can stop spending energy on shame and start spending it on the plan.

Here’s the part I couldn’t fit into the video.

The people who got through the last few funding shocks all had one thing in common. They were never dependent on a single source in the first place.

I’ve seen a few of these cycles now. Money comes in, it dries up, then it turns up somewhere else. And the ones who survive usually aren’t the biggest labs or the cleverest people. They’re the ones who had three small streams instead of one big one. A national grant. A society travel award. A little teaching or consulting money covering a few months. None of it impressive on its own. Together, it’s very hard to switch off with one email.

Put it in terms we all use: you’d never build a conclusion on a single data point. We’d all call that shaky. But most of us build an entire lab on one funder and call it a plan. That’s really what this cut exposed; a lot of careers were running on one number with nothing behind it.

So here’s the one thing I’ll ask you to do this week, and it’s not in the video.

Work out your funding concentration, as a plain percentage. In the last two years, what share of your lab’s money, or your stipend, if you’re a student, came from one source? If it’s 80 or 90% from a single place, that’s not a failing on your part. Almost everyone I know in our part of the world is in the same spot. But now you have the number. And from here the job is simple to say, if not easy to do: bring it down. Not by landing one new giant grant, but by adding small, unglamorous, steady streams until no single cut can end you.

Let me be concrete about what those streams can actually be, because “diversify” is useless advice without a list. National and regional science foundations in your own country, which are often underused because they’re less prestigious than the big international names. Scientific-society grants and travel awards, which are small but reliable and build your reputation as a bonus. Bilateral programs between your country and a partner one. Philanthropic and disease-focused foundations, many of which are actively looking to fill the gap the aid cuts left. Modest industry partnerships tied to a local problem. And yes, teaching or consulting income, which nobody puts on a CV but which has quietly kept more labs alive through a bad year than any of us admit. You will not replace one $500,000 grant with one equivalent. You replace it with five or six smaller things that, together, cannot all be cancelled by the same signature.

There’s a second habit worth starting now, while the shock is fresh enough to motivate you: keep a simple one-page “funding map.” Every source you’ve ever applied to, every one you’ve heard a colleague mention, every deadline you know of, in one document. Most of us carry this in our heads, which means it vanishes the moment we’re stressed, exactly when we need it. On paper, it becomes a plan you can work through calmly instead of a panic you improvise.

And do one more thing while you’re at it: talk to your own research office or finance person before you assume the worst. In the panic, people walk away from money that was still theirs. Ask the boring questions: is there a no-cost extension, can unspent funds carry over, is any of it merely paused rather than cancelled. A lot of the freezes came with a twelve-month pause attached, not a full stop, and some of those were later reversed. Don’t return a cheque you didn’t have to.

One last thing, because I don’t want to leave you feeling flattened. Even the NIH didn’t ban working across borders. It changed the mechanism: direct, independent awards now instead of the old sub-award. Some grants that looked dead, including several to South African scientists, were brought back. The money didn’t leave science. It moved. And in the next video I’ll show you where it went: Europe’s push to pull in scientists, the Gulf, and the South-to-South funding that almost nobody tells you about.

Do me a favor: hit reply and tell me your concentration number. Just the percentage, even if it’s an ugly one. Saying it to one person is sometimes what makes you finally do something about it. I read every reply.

Talk soon, John

Where the numbers come from: reporting on the 2025 US foreign-aid review (USAID and State), NIH policy notices and the 2026 national researcher survey, and coverage in Science, STAT, and Devex.

If someone forwarded this to you and you want it every week, you can subscribe here: [researchroadmap.substack.com]

I want to tell you about a student I met a couple of years ago, not long after I moved to Sharjah.

Bright kid. Hardworking. He came into my office and proudly told me he was going to use a fancy new sequencing method for his project. I asked him one question: “Okay, but what question are you actually trying to answer?” He went quiet. He didn’t have one. He had a method he was excited about, and he’d spent eight months building a project backwards from it.

That’s the mistake. And it’s so common that I made this week’s video about it. In the video I walk through the three filters I use to choose what to work on, even when the lab is small and the budget is basically nothing. I won’t repeat all of it here. But I want to give you something I didn’t put in the video, because you’re on this list and you deserve a little extra.

Watch the video here:

In the video I walk through the three filters I use to choose what to work on, even when the lab is small and the budget is basically nothing. I won’t repeat all of it here. But I want to give you something I didn’t put in the video, because you’re on this list and you deserve a little extra.

The “two emails” test

Here’s a trick I use before I commit to any project. I imagine I’ve finished it. The paper is published. Then I ask myself: who are the two people who would email me about it?

If I can picture two real, specific people, by name, in my field or next to it, who would read that paper and care, then the question is probably worth doing. A farmer in my region. A colleague studying the same soil. A policymaker who needs the number.

If I can’t name two people, that’s a red flag. It usually means I’ve fallen in love with a technique, or I’m chasing something that’s interesting only to me. Interesting-to-me is not the same as important. I’ve learned that the hard way, more than once, over thirty years.

The beauty of this test is that it costs nothing and takes thirty seconds. You don’t need funding to do it. You don’t need a big lab. You just need to be honest with yourself before you spend two years finding out the answer the expensive way.

One thing to try this week

Take whatever project you’re considering right now. Write down the names of the two people who would email you about the finished paper. Actual names.

If you get stuck, that’s not failure. That’s the filter doing its job, and it just saved you a year.

That student, by the way? We sat down, threw out the method, and started from a question that mattered in his home country. His project got smaller, cheaper, and far better. He’s publishing now.

Reply and tell me the two names you wrote down. I read every email.

Talk soon, John

P.S. If a colleague is about to start a project they haven’t pressure-tested, forward them this. It might save them a year too.

I want to tell you about the project that saved my independence, precisely because it was the one nobody else wanted.

Early in my career, most of my time went into other people’s questions. That’s normal, and it isn’t a complaint. That’s how you learn, and some of those collaborations were the best training I ever got. But there was one small thing I worked on in the cracks. A question about soil organisms that nobody senior thought was worth much attention. No one fought me for it, because no one wanted it. It was, quietly, entirely mine. The question sounds small - Do soil invertebrates harm arbuscular mycorrhizal symbioses in the field? For a while it felt like the least important thing on my plate. Looking back, it was the most important. Because when people thought of that question, they thought of me. Not me-plus-a-famous-name. Just me. It became the thing that made me a researcher with my own identity, instead of a very capable pair of hands attached to other people’s labs.

That’s the part I couldn’t fit into this week’s video, so I'm sharing it here.

Keep one project that is wholly, unmistakably your own.

The video (see below) is about protecting your voice inside a collaboration, having the authorship conversation early, owning a visible piece, learning to disagree without burning the relationship. All of that matters. But there’s a deeper kind of insurance the video doesn’t get to, and it’s this: the surest way to never be swallowed by a senior collaborator is to never be entirely dependent on them in the first place.

Most early-career researchers I meet, especially those working far from the big centers, have the opposite setup. Every project they touch belongs, in some real way, to someone more powerful. Someone else’s grant, someone else’s question, someone else’s lab. And when all of your work is shared, you have no leverage and no separate identity. If a collaboration sours, you don’t just lose a project, you lose everything you were working on at once. That is a frightening position to negotiate from, and senior people can feel it.

So carve out a sliver that is only yours. It can be small. It almost should be, at first. A side analysis. A question your supervisor finds mildly interesting but not exciting. A dataset you can get to that nobody else can be bothered with. Protect maybe ten or twenty percent of your time for it, and guard that time the way you’d guard money. The point isn’t the size of the project. The point is that one line of work exists where you are the first author, the corresponding author, the person the field eventually associates with the idea.

Here’s why this changes everything, quietly:

It gives you an identity that travels. When you apply for the next position, the fellowship, the grant, reviewers want to see a thread that is yours, evidence you can ask your own questions, not just execute someone else’s. A CV full of middle-author papers on important work still reads as “good helper.” One independent line, however modest, reads as “future leader.” That distinction decides careers.

And it changes how you sit in every other room. When you walk into a collaboration knowing you have something of your own, you negotiate differently. You’re not clinging. You can be generous, easy, genuinely helpful, and able to say no, because not everything is riding on this one senior person’s goodwill. Independence isn’t a wall you put up. It’s a quiet floor you stand on, so you can afford to be open.

I’ll be honest about why I keep pushing this. For a long time I measured my worth by whose projects I’d been allowed into, as if access to important people’s work was the same as having my own. It isn’t. The work I’m proudest of, the work that actually became mine, started as the unglamorous side project nobody competed for. I almost didn’t bother with it. I’m so glad I did.

So this week’s assignment is just one question, written down honestly: what is the one project that is truly yours? If you can name it instantly, good! Protect its time fiercely this month. If you can’t name one at all, that’s not a failure. That’s the most useful thing you could have discovered today. Start it small. Start it this week.

And do me a favour, hit reply and tell me what your own project is, or what it would be if you gave yourself permission. Sometimes saying it to one person is what makes it real. I read every one.

Talk soon, John

Hi — good to have you back!

I want to tell you about a man whose name I’m not sure I could have spelled correctly at the time.

Years ago, early in my career, still anxious, still convinced everyone in the room was smarter than me — I gave a small, forgettable talk at a workshop. Maybe fifteen people in the room. Afterward, one of them came up, asked a couple of sharp questions, and we talked for ten minutes by the coffee. That was it. We didn’t become friends. We barely even emailed. He drifted off to the edge of my world, and I forgot about him.

About four years later, an opportunity landed in my lap, the kind that genuinely changes the shape of a career. When I asked how my name had come up, the answer was: that man. The one from the coffee. He’d remembered something I said in those ten minutes, and when someone asked him “who works on this?”, he said me.

I’ve thought about that a lot since. Because it points to one of the most useful, and most hopeful, things I know about how careers actually work.

Your biggest opportunities will almost never come from your closest circle.

This isn’t just my experience; it’s a well-studied pattern. The people closest to you, your lab, your department, your usual collaborators, tend to know the same things you know and the same people you know. They’re a warm, comfortable echo. The genuinely new opportunity, the grant you’d never heard of, the collaboration on another continent, the invitation out of nowhere, almost always arrives across a weak tie. An acquaintance. Someone at the edge.

For a researcher who feels isolated, far from the famous clusters, this should land like good news. Because it means you don’t have to somehow force your way into a dense group of stars. You don’t need to be in the in-crowd. You need bridges, a scattering of loose connections reaching outward into worlds you’re not part of yet. And bridges are far easier to build than memberships.

The video this week walks through how to build those connections from scratch — the give-first move, the one good email, all of it — so I won’t repeat that here. What I want to give you instead is the part almost nobody does, and that costs you nothing: not making new contacts, but reviving the ones you already let go cold.

Here’s what I mean. Right now, scattered through your memory and your inbox, you have dozens of these — someone you met once at a conference, a co-author from a project that ended, a person who sent you a kind question two years ago that you never quite answered. We treat these as dead. They’re not dead. They’re dormant. And a dormant tie is the single most underused asset in a researcher’s career, because all the awkward work of meeting has already been done.

So this week, instead of cold-emailing a stranger, do something easier and stronger: reach back to one person you’ve lost touch with.

And do it without an ask. That’s the whole trick. Don’t reconnect because you need something; people can smell that, and it turns the eventual request into the only reason you ever wrote. Reconnect because you genuinely remembered them. Three lines is plenty: what you remembered about them, what you’ve been working on, and a small gift if you have one: “I saw this paper and thought of your work,” or “your method ended up shaping how I approached this, thank you.” No request. Just put yourself back on their radar as a real person doing real work.

Some of those threads will pick right back up. And the next time that person is standing in a room where someone asks “who works on this?” - you’re a name they can reach for again.

I’ll be honest about why I keep hammering on this. For thirty years I quietly assumed the people getting all the opportunities were simply better. It took me an embarrassingly long time to see how much of it was just connection — who remembered them, who thought of them, who vouched for them in a room they weren’t in. Almost none of it was some secret brilliance. It was a web. And a web is buildable, from anywhere, by anyone willing to be useful and to stay in touch. That’s a fairer story than the one I used to believe — and it happens to be the true one.

So this week: one dormant tie. One genuine, no-strings message. That’s the whole assignment.

And do me a favor — hit reply and tell me about a connection you’ve been meaning to rekindle but haven’t gotten around to. Sometimes saying it to me is the nudge that finally gets you to send it. I read every one.

Talk soon,

John

I want to start with a confession, because it’s the thing that took me the longest to admit, and it’s the reason this newsletter exists at all.

For a big chunk of my career, I was a reviewer. And for a few of those years, an editor. Which means I was one of the people sending the emails. The “we regret to inform you” emails. I’ve probably written some version of that line a few hundred times.

So let me tell you what was actually going on when I had your paper in front of me — because I think it’ll change how you submit your next one.

It usually went like this. I’d have a stack of papers to get through, and not nearly enough time for any of them. So I’d open one up, and I’d read the title. Then the abstract. Then the first paragraph or two of the intro. And somewhere in those first few minutes — and I’m not proud of this — I’d already made up my mind about whether I cared.

Not whether the science was good. I hadn’t even gotten to the methods yet. I’m talking about whether I cared. Whether the paper had given me a reason to keep going.

That’s the part nobody tells you. The decision often gets made before anyone reads your data.

And here’s the thing that bothered me later, once I’d left North America and started working with researchers in other parts of the world. The papers I’d been quickest to lose patience with weren’t the worst science. A lot of the time they were really good science. They were just written in a way that buried the point. The author would spend the first page proving how careful they’d been — every control, every method, every reference to the foundational literature — before ever telling me why I should sit up in my chair.

I get why people do that. If you’ve spent your whole career being made to feel like an outsider, of course, you want to prove you belong. Of course, you front-load the rigour. But the people on my side of the table read that as “this person doesn’t know what the headline is.” And we’d move on.

Meanwhile, a researcher from some famous institution would open with one confident sentence about why their work mattered, and we’d give them the benefit of the doubt on everything else. Same quality of science. Completely different reception. I watched it happen for years before I let myself really see it.

I’m telling you this because I don’t want you to take your next rejection personally. When you get “lacks novelty” or “insufficient impact,” what’s usually hiding underneath is something much more boring and much more fixable: I didn’t understand why I should care fast enough. That’s not a science problem. That’s a first-three-paragraphs problem.

Now, this week's video walks through the three big mistakes I see over and over, so I won’t repeat all of that here. If you’ve watched it, you’ve got the framework. What I want to give you, as someone who took the time to subscribe, is the stuff I couldn’t fit on camera. The practical part. The “okay, John, but what do I actually do Monday morning” part.

So here’s what I’d do this week if I were you.

Pick one paper. Just one. Something published in the last year or two, in the journal you’re aiming for, in your field, that got in. Now read it — but not for the science. Read it like a spy. How does the very first sentence work? How long before they start talking about their own results? Watch how they move from “here’s our specific little finding” to “here’s why this matters for the whole field.” That move, that zoom-out, is the thing most rejected papers never make. Once you start noticing it, you can’t unsee it.

Then go back to your own abstract — the one that got rejected, or the one you’re about to send — and be honest about whether it leads with the point or buries it. Most of us bury it. We warm up for three sentences before we say anything. Cut the warm-up. Open with the thing you’d say if you only had one sentence to make a stranger care.

And read your discussion section one more time with a slightly uncomfortable question in mind: am I actually arguing for why this matters, or am I just politely describing what I found? There’s a real difference, and a lot of us — especially those of us who were raised to be modest — only ever do the second one. Your discussion is allowed to make an argument. It’s supposed to.

One more, and this is the one people skip: if you got rejected, don’t just bundle the paper up and fire it off to the next journal unchanged. I know the temptation. I’ve felt it. But sit with the feedback for a day, even when it’s vague and annoying and seems unfair. There’s almost always something real underneath it. Find someone who knows your field and ask them to help you translate what the reviewer was really reacting to. Reviewers are bad at saying what they mean. They’re rarely wrong that something didn’t land.

I’ll be honest with you about why I’m doing this whole thing — the channel, this newsletter, all of it. I spent thirty years inside a system that I now realize quietly advantages some people and quietly disadvantages others, and for most of that time I couldn’t see it because I was one of the advantaged ones. I can see it now. And it turns out almost none of it is about talent. It’s about knowing the unwritten rules. So I’m just going to spend my time handing those rules to the people who never got them.

That’s what you’ll get from me here, more or less every week. Some of it is tied to a video, a lot of it is not. Stories from the inside, practical stuff you can use, and the occasional rant.

If you’ve got a paper sitting in a drawer because it got rejected and it stung — go dig it out. It’s probably a lot closer than you think. Usually, it’s one honest revision away from landing somewhere good.

And do me a favor: hit reply and tell me what you’re stuck on right now. I read all of them, and honestly, it’s the best way for me to figure out what to write about next.

Talk soon, John