It’s The Crawl – Sunday, Aug. 30. Five things worth reading in public relations, communications and AI search as you head into the week.

1️⃣ “Show more” shows itself out

Google is expanding some AI Overviews to full length on its own, with no “Show more” click required. Chris Long, co-founder at Nectiv, spotted the behavior and posted it on X, and Search Engine Land’s Barry Schwartz got Google to confirm it on Aug. 28.

The expansion pushes the core search results down the page, and the result page starts to look like AI Mode. Google did not say how many queries it covers.

Google the questions people ask about your brand and watch which answers open to full length on their own. Your screenshots are the only record of it you’re going to get.

Source: Read more at Search Engine Land.

2️⃣ Where the answer comes from

Shero Commerce, a Shopify agency that sells SEO and AI services, ran shopper buying questions across 60 product categories through Google AI Mode, ChatGPT and Perplexity in July 2026, then counted the 1,851 sources those answers cited.

2.8% of them were a brand’s own domain. 59% were publishers, the study’s term for third-party editorial sites, and its own examples run from Forbes to Good Housekeeping.

Build your media list for that mix.

Source: Read more at Search Engine Journal.

3️⃣ Written for two readers now

Digiday reported that USA Today Co. is testing ways to reformat its articles for humans and for machines at the same time, including converting webpages into markdown.

The company says it already blocks about 99% of the AI bots that declare themselves in their user agent, and whitelists only the ones it has approved relationships with. Access became the thing it sells.

Check the outlets on your target list to see if they disclose a crawler policy. It might impact your media list if you’re trying to cover all your AI visibility bases.

Source: Read more at Digiday.

4️⃣ Google’s forum block has a regular

Ahrefs analyzed 145 million U.S. Google search results and found the “Discussions and forums” block, a panel of forum links Google places on the search results page, on 11.7% of them.

When the block appeared, it carried a Reddit link 83.9% of the time worldwide. Ahrefs does not say what period the data covers.

Reddit’s place in this block says nothing about its place in an AI answer. That is a separate surface with its own numbers. Check which one a Reddit figure came from before you put it in an email or deck.

Source: Read more at Ahrefs.

5️⃣ Everyone’s favorite, sorted by age

YouGov asked U.S. adults who had used generative AI in the past 30 days to pick the tool they like using most, out of the ones they would consider using at all.

ChatGPT leads every generation, from 44.4% of Gen Z down to 24.5% of Baby Boomers+. Claude is the pick of 10.6% of Gen Z and 1.4% of Boomers+, the steepest generational split in the set. The data was collected from February to July 2026.

Which AI platform someone likes is a fact about your audience. It says nothing about how any of them describes you. If the people you’re trying to reach skew younger, check what more than one platform says about you. The platform you default to may not be the one they’re using.

Source: Read more at Search Engine Journal.

Subscribe now

The companies that spent the year chasing Reddit citations learned that lesson the hard way.

Some paid agencies to post about them in subreddits, creating fake posts made to look like real user reviews.

On Aug. 14, the ground shifted. (poof!)

Reddit’s share of ChatGPT citations collapsed 86.4%, first reported by Promptwatch.

It’s not the first time Reddit’s citation share has swung, it won’t be the last. Forbes and Axios later published their own takes.

For comms, the why shouldn’t matter here.

The drop is a warning about how companies are approaching AI visibility, and the importance of building a resilient narrative.

What you’ll get in the next 5 minutes

• Why concentrated authority is the risk

• What GEO is turning into for comms teams

• How to build a narrative with no single point of failure

• Two checks you can run this week

Subscribe now

Concentrated authority is the risk (not Reddit)

It’s easy to understand why Reddit became so appealing to brands.

Media coverage from some of the biggest ad trades, webinars and workshops on building real community to drive visibility and sales, countless ‘how to’ guides for building a presence.

Brands saw Reddit, a trusted source for community opinion, being cited heavily and started treating visibility on the platform as an AI-search tactic.

AI search rewards third-party corroboration and Reddit presented an opportunity to manufacture it, quickly. The gold rush ensued (and was well documented).

Marketers deployed AI-generated “stealth marketing content” to create artificial hype, fake reviews and fake votes, according to MediaPost.

Reddit eventually cracked down.

In July, Reddit said it was catching about 25,000 net new spammy posts and comments every day and had revoked almost two million inauthentic votes per day over the last three months.

The response now from some brands will be to leave Reddit and chase the next citation winner.

For serious comms teams, that’s not a strategy. All external authority is borrowed and Reddit will continue to be one door AI checks.

The danger is when the authority you’re borrowing is actually concentrated authority, and you’re dependent on one platform to carry the entire weight of your narrative.

Reddit was that platform for a lot of brands, so when its weighting changed, their narrative changed with it.

For comms teams, GEO is becoming a reputation strategy

Stop relying on one platform to carry your story and GEO becomes reputation strategy.

AI search applies traditional communications principles under harsher conditions.

For example, a journalist examines a handful of sources before filing a story. But an AI system reassembles a company’s story continually, from different sets of sources, for differently worded questions, all day.

Every answer is an automated reputation test. Does your narrative remain coherent when AI rebuilds it out of whatever evidence is available at that moment?

The Reddit shift changed the mix of available evidence and created a surprise stress test for every company narrative built on Reddit.

When you begin thinking of GEO as reputation strategy, it becomes about how you increase the likelihood that the same credible understanding emerges regardless of which sources AI selected.

A Reddit citation is one route. Other routes include an executive interview, corporate explainer, original research or an analyst report.

You’re building a case and that’s work comms teams already know how to do.

Is the position clear? Can we support it with evidence? Do credible outsiders validate it? Does our own language match the public record? Will the story stay persuasive when it is retold without us?

The result is narrative resilience: the point where the surrounding evidence is strong enough that AI can’t reasonably conclude anything else about your company.

Subscribe now

A strong narrative has no single point of failure

When a company builds a resilient narrative, it builds a world that has no single point of failure.

Every layer does a different job.

• Your own site carries the first-party record.

• Independent coverage examines it.

• Executives carry the ideas in their own voice.

• Original research gives other people something to cite.

• Analysts supply category context.

• Communities are where customers describe what happened to them.

Consistent language across all six is what lets AI easily connect them.

No AI platform will use every layer of your narrative. They shouldn’t need to. Each form of evidence points toward the same understanding on its own.

There is a test for this (and it’s easy!).

Take Reddit out. Does your company’s narrative still hold? Take the corporate site out or the biggest media placement out. Change the AI platform or the model.

If the answer collapses with any layer removed, you have visibility but not resilience.

Two checks you can run this week

• Run that test on your live answers. Ask ChatGPT and one other engine the five questions a customer or stakeholder would actually ask, save the source list under every answer, and mark the one domain that would take your story with it if it disappeared. That mark is your concentration risk.

• Map the claims your narrative depends on. For your two or three most important claims, write down where the first-party record sits, who independently validates it, and which person carries it in public. A missing answer means there’s work to do and a claim that rests entirely on one platform or source is at risk.

Does it hold up?

Reddit’s citation share on ChatGPT will probably bounce back. The collapse has stabilized, though at the lower level, according to Promptwatch. Time will tell.

None of that changes the job.

The best comms teams will not try to predict every platform, model or weighting change.

They will build a world around the company’s narrative: clear enough to understand, evidenced enough to believe and distributed widely enough to survive any one door closing.

Subscribe now

It’s The Crawl – Sunday (night), Aug. 23. Five things worth reading in public relations, communications and AI search as you head into the week.

1️⃣ YouTube chapter markers are comms work now

BrightEdge’s research across nine industries, from May 17 to Aug. 2, 2026, showed that when Google AI Overviews cites a YouTube video, half the time the link goes to a specific timestamped moment in it. ChatGPT’s YouTube citations almost never do.

One video has to work for both, which turns chapters, transcripts and spoken transitions into part of your narrative infrastructure rather than production polish.

Whoever uploads the video is making that decision now, and probably has no idea. The easiest win on BrightEdge’s own list is an accurate uploaded transcript instead of auto captions.

Source: Read more at BrightEdge.

2️⃣ Local businesses carry the local record

Steady Demand ran 1,487 local service queries across 50 U.S. metros in late July 2026, through the Gemini API with Google Search grounding, and found nearly 60% of what Gemini cites about a local business comes from that business’s own pages.

Ask the same question minutes apart and the top business it names changes roughly nine times in ten.

So check it more than once before you report anything, and spend time on the pages you actually control. Ben Fisher, who ran the study, calls the instability Grounding Drift.

Source: Read more at Search Engine Land.

3️⃣ It reads about 1 in 80

RESONEO analyzed 61,332 URLs that ChatGPT surfaced in its sources sidebar across 1,249 captured conversations in July 2026 and found it opened and read 759 of them, all but two in Thinking mode.

A page it opened in Thinking mode was cited 74% of the time. A page it retrieved from memory and never opened, 7%.

An opened page is roughly ten times likelier to be cited. So ask whether your AI visibility report counts pages that appeared in the sources sidebar or pages ChatGPT actually read. They are different numbers.

Source: Read more at Search Engine Land.

4️⃣ Word order moves the answer

Google Research quizzed 13 AI models on 2,150 facts taken from Wikipedia. Two of the newest models in the test, Gemini 3 Pro and GPT-5, hold 95 to 98% of those facts and still fail to recall 26 to 34% of them when asked directly.

One reason is order. A model learns a fact during training with its subject and object in a particular sequence, and a question that flips that sequence is harder for it to answer.

None of it was measured on brands, and Google offers no advice about writing, though Search Engine Journal’s write-up floats one: order the entities the way people usually ask.

Source: Read more at Search Engine Journal.

5️⃣ Creators formalize AI discovery

Digiday reports independent creators treating AI citation as a discovery channel, auditing where AI search quotes them and leaving verified crawlers unblocked.

The article is based on creator anecdotes. One creator, Courtney Johnson, said her own Google Search impressions went from 2,568 in June to 3,048 in July, with nothing tying the change to AI. Another creator, Colin Rocker, audited his own citations with Claude and found he gets quoted through his LinkedIn profile and his podcast appearances more than through his own site.

Before your next partnership call with a creator, decide the metrics and methodology around AI visibility that clear your bar.

Source: Read more at Digiday.

For years, it’s been a home for everyone. Vague language, built to accommodate every stakeholder, updated when an executive bio changes.

Well, that’s over.

The corporate site is now a narrative hub: a current and coherent account of a company’s facts, ideas and evidence. Built so people and AI platforms can understand how it all fits together.

Companies are already staffing to build one.

That brings new tensions, new decisions and a public test of the story you’ve been telling.

What you’ll get in the next six minutes

• Why more of your company’s story now comes from what you publish

• What a narrative hub is, and who companies are hiring to build one

• Why having a strategy isn’t the same as having a record of one

• Three checks to run on your own site

Subscribe now

The press corps shrank while AI search grew.

Two big shifts are happening in the world communications and PR teams work in.

The first has been building for years. The people whose job it was to write the outside record thinned out, and the pressure on the ones who stayed grew.

There are roughly 17,000 fewer U.S. news analysts, reporters and journalists than there were in 2000, more than a quarter of the profession, by the Bureau of Labor Statistics count the Wall Street Journal carried.

The reporters who remain are stretched thin. 49% name shrinking budgets, staff cuts and heavier workloads as their biggest obstacle, in Cision’s 2026 State of the Media Report, up from 29% who said the same a year earlier.

So they turn to PR teams. 66% rely on PR-provided content for story ideas, which Cision reports as the leading source of leads, and what they say they want inside it is evidence: original research, access to experts, embargoed material.

The second shift is AI search, and how fast people moved to it. 42% of U.S. adults now say they use chatbots to search for information, the most common use reported in Pew Research Center‘s survey. A majority say they read the AI summary at the top of a search result.

Journalists moved too. Cision found 43% of them use generative AI for research or fact-checking, and 48% for brainstorming story angles and interview questions.

When AI answers, it pulls from the public record that exists, including what’s on your corporate website.

What used to influence visitors now has to influence answers, and what that demands of companies and their comms teams is different.

(The journalist counts are the Bureau of Labor Statistics’, reported by the Journal rather than produced by it. Cision surveyed 1,899 journalists across 19 markets in January and February 2026, and sells PR software. Pew surveyed 5,119 U.S. adults that same February.)

A content hub accumulates content. A narrative hub accumulates understanding.

The corporate site’s influence now reaches well past the people who visit it.

Your corporate site is your company’s primary contribution to the public record. Where your narrative becomes public, retrievable and testable. And where you put down the source material AI uses to construct a narrative of you.

A former colleague described it to me as a knowledge system, which is a great way to think about it. The place where AI retrieves a current record of the company, pulls first-party evidence out of it, and can arrive at an accurate read of your strategy and your position in the market.

None of that means AI will believe everything you publish. It gives AI a clear version of your story to test against everything else online. Without it, the record is written by someone else.

That requires a shift in how companies think about the site, from traditional content hubs to narrative hubs.

The companies acting on it are building narrative hubs, and hiring storytellers, editors-in-chief and heads of narrative.

The share of U.S. LinkedIn job postings containing “storyteller” doubled in the year to Nov. 26, 2025, by LinkedIn’s own count in the Wall Street Journal.

Executives said “storytelling” on earnings calls and investor days 469 times in 2025 through Dec. 11, against 147 in 2015, by FactSet’s count in the same piece.

Visa has been staffing a newsroom. A job posting live until Nov. 21, 2025 sought a VP of Global Brand Communications and Newsroom, to establish a new function and put a centralized editorial calendar and a governance model behind it.

They also posted roles for an Editor-in-Chief and News Director earlier this year.

Look across the titles posted by other companies since. Head of Narratives. Head of Corporate Narrative and Messaging. Head of GTM Narrative. VP of Brand and Narrative.

Underneath all of it, the same goal: building a narrative consistent and durable enough to hold together wherever, whenever and by whomever it is read. Today, that means holding up across AI platforms, new models and weighting changes.

This is infrastructure, and comms teams are the ones being asked to build it.

(Both counts are LinkedIn’s and FactSet’s, reported by the Journal rather than produced by it.)

Subscribe now

AI turned the corporate site into a public test of strategic clarity.

AI rewards companies that can explain themselves clearly, and companies are finding that having a strategy is not the same as having an intelligible public record of that strategy.

Corporate communications may own reputation. Brand may own positioning. Marketing may own customer content. Investor relations may own the financial story. Executives may have their own platforms.

Without a connecting role, each group can publish a valid but different version of the company. Hence the hiring of heads of narrative and infrastructure build-out.

This shift invites hard choices.

How explicit do we get publicly? Where does our record sit now and where should it sit? Who is the steward of our narrative across all owned channels?

Getting explicit costs something real.

It means committing to one vocabulary, spelling out the market logic behind an inspirational vision statement, producing first-party evidence to prove your claims and defining the categories you say you play in.

Corporate sites were written to accommodate everyone, so the language got broad enough that every function could see itself in it.

“A purpose-driven, customer-focused company using innovation to create meaningful experiences.”

“We are transforming the customer experience.”

“We are building the future of our industry.”

That vagueness was deliberate. It preserved flexibility, and human readers filled the gaps with what they already knew about the brand.

But ambiguity in an AI environment results in an empty narrative. It leaves AI to infer the answers from the rest of your public record, and leaves your own site with nothing specific enough to quote.

And for many companies, the public record exists across more than one site.

About and newsroom pages may sit on the main domain or on a corporate subdomain of it.

Investor relations usually sits somewhere else again: of the large U.S. public companies I checked that have a dedicated investor-relations subdomain, roughly two-thirds serve it from a third-party investor-relations platform, and a single vendor runs 31 of them.

None of those setups are wrong, and AI is perfectly capable of crawling subdomains. But with nothing connecting them, the version it assembles can be one you don’t recognize.

Which leaves an uncomfortable fact: no single traditional function fully owns the multitude of ways AI constructs a narrative about your company.

What we’re seeing now is the beginning of how companies fix that.

(The vendor sweep covered 101 large U.S. public companies on Aug. 12, 2026, a roster assembled for the test rather than an official list, and looked for four subdomain naming patterns. Another 23 resolved none of them, which is not evidence they self-host.)

Where to start

Three checks, in order.

1. Read your About page, your newsroom and your investor relations page side by side. Ask whether they describe the same company, in the same or similar words, doing the same thing.

2. Look at how your corporate site is organized. Are you publishing content, or building a narrative someone could piece together no matter which page they land on?

3. Name who owns the narrative across every owned channel. Comms, brand, marketing, investor relations and your executives all publish. If no single name comes back, you have found the thing to fix.

Does it hold up?

Building a narrative hub is expensive. It requires a new level of maintenance. Most importantly, it’s a long-term play. The companies funding editors, daily editorial standups and governance models are not doing it because a study told them to.

They know their company’s record now is written long before someone asks about it.

The record has to work for someone who arrives at the front door and reads in order, and now for a reader who never comes in that way.

Subscribe now

It’s The Crawl – Monday, Aug. 17. To start your week: four things worth reading in public relations, communications and AI search.

1️⃣ LinkedIn had a month

Meltwater’s July 2026 analysis across eight models found LinkedIn’s citation volume up 67.8% month over month, into the top five cited sources.

Earned media and news hold 33.1% of citations.

Social reached 7.8% of all citations, its highest in the five months Meltwater has tracked it, while Wikipedia, NIH and Statista all fell.

The pull is toward LinkedIn specifically, so any executive thought leadership plan has to treat it as a primary channel, not a place to cross-post the blog.

Source: Read more at Meltwater.

2️⃣ Reddit overtakes Wikipedia in ChatGPT

BrightEdge tracked citations across Google AI Overviews and ChatGPT from February to July 2026 and puts Reddit’s growth at about 84%, against 37% for Wikipedia.

Reddit passed Wikipedia inside ChatGPT for the first time in May, though ChatGPT is the smaller of the two engines by volume. AI Overviews carries roughly 88% of Reddit’s citations. BrightEdge does not disclose how many queries the sample covers.

Go read what your category’s subreddit says about you. Reddit is one of the doors AI opens when answering questions about your company, and most teams have never looked at what’s behind it.

Source: Read more at BrightEdge.

3️⃣ The user-generated web wins

Kevin Indig and Amanda Johnson analyzed about 35,000 ChatGPT citations from December 2025, all of them SaaS-related prompts.

User-generated content (UGC) platforms take 17.1% of cited domains against 4.0% for publishers, and Wikipedia, Reddit and LinkedIn are 99% of those UGC citations. Three addresses, and most earned media plans name none of them.

Put a name against each of the three this quarter. The Wikipedia entry alone needs an edit-request plan and someone who knows how those actually work.

Source: Read more at Growth Memo.

4️⃣ LinkedIn wrote the manual

LinkedIn published its own AI search playbook on August 11, built on Profound’s March 2026 finding that LinkedIn is the most-cited domain for professional queries.

It leans on a May 2026 Corporate Ink survey where 34% of U.S. B2B tech marketers say they have an AI visibility strategy they feel ready to run, and 51% name limited in-house knowledge as the obstacle.

The profiles this playbook depends on belong to individual executives, not your brand page. The CEO owns the category narrative and the company vision, the leadership team deepens it with their own expertise, and that is where an executive thought leadership plan starts.

Source: Read more at LinkedIn.

It’s The Crawl – Sunday (night), Aug. 9. Three things worth reading in public relations, communications and AI as you head into the week.

1️⃣ Crawl volume without the traffic

On roughly 200 retail sites Botify analyzed earlier in 2026, OpenAI read 198 times for every person it sent. Google’s ratio was one in six.

AI bot traffic on those sites grew 5.4x during 2025.

OpenAI reads a page 198 times for every person it sends there, so read one of yours as plain text and check whether it still answers the question on its own.

Source: Read more at Botify.

2️⃣ Zero click has an age gap

Harry Clarkson-Bennett cites iPullRank’s July 2026 analysis of Datos clickstream data (October 2024 – December 2025), where 52.3% of searches by 18- to 24-year-olds ended without a click against 42.7% for 55- to 64-year-olds. The gap comes from one week of panel data.

The audience still clicking through is the oldest one you have.

Younger audiences meet you inside the answer and never on your page, so what AI says about you is the whole impression for them.

Source: Read more at Search Engine Journal.

3️⃣ Far from home, you’re a footnote

Kevin Indig of Growth Memo analyzed Semrush’s U.S. ChatGPT data across 1,094 categories from January to June 2026.

In the categories a brand is known for, 74% of its appearances cite it and 44% name it. In distant ones, citations hold near 50% while names fall to 25%.

Expanding into a new category earns you a citation long before it earns you a mention by name, so expect a stretch where AI uses your evidence without saying whose it is.

Source: Read more at Growth Memo.

Two words are nowhere in it: earned media.

That should concern anyone who owns a brand’s story, especially the person signing off on the budget behind it.

An AI answer gets assembled in seconds out of whatever the model went and found, or remembered, and most of that wasn’t written by you or your company.

What you’ll get in the next six minutes

• What IAB’s framework measures, and what it gets right

• Why earned media belongs in it, and the metric that would measure it

• Why AI reasoning modes belong in it too

• Two moves, starting with a count you can run in an hour

Subscribe now

What does IAB’s AI visibility framework measure?

More than 20 companies now sell tools that tell you how you show up in AI answers, and two of them can look at the same brand in the same week and hand you different numbers.

That’s the Interactive Advertising Bureau’s (IAB) own starting point, and Measuring Visibility in the AI Era is a serious attempt at fixing it.

IAB calls its four measures the 4 P’s of AI visibility: Presence, Prominence, Portrayal, Persuasion.

Underneath the labels, four questions. Does AI mention you at all. How high up and how visibly. In what light, and how accurately. And does it recommend you, or just list you alongside everyone else.

Then the framework does the harder thing and says what counts as good enough. A tool running a handful of prompts gives you a hunch, not a measurement. And it tells you what to demand from a vendor, plus the best line in the document: if they can’t or won’t answer, treat the silence as the answer.

IAB’s focus is the media and marketing industries, so a framework that doesn’t center PR and earned media is no surprise. Its measurement standards have almost always governed media somebody bought.

This framework is different. It measures visibility nobody bought, which as far as I can find IAB has never done at this scale.

I give IAB a lot of credit.

This is real work for a channel changing under everyone’s feet, and their reason is right: optimization without measurement is guesswork, and a brand can’t know whether its GEO investment is working unless everyone measures the same way.

But one large thing is missing from their framework, and it’s one of the biggest inputs to AI visibility.

(IAB did publish a narrow 2015 guideline on social platform engagement and sentiment, but that was 10 years ago.)

Earned media shapes AI answers. The framework doesn’t measure it.

Muck Rack has been counting the links AI cites, and across more than 25 million of them from ChatGPT, Claude and Gemini in 17 industries, journalism alone is about 27% of everything cited.

That figure has barely moved across three counts in ten months. Goodie AI, across 45 million citations, and a University of Toronto team both put the whole earned category higher still. They disagree about the size. Not one has it small.

When a buyer asks AI about your company, it pulls what it can find and writes the answer from that.

If your site says you lead the category and three trade stories call you the scrappy alternative, the answer is built from both, and you only wrote one of them. Your authority and positioning get written down by other people, then read back to your buyer as fact.

Now look at what a brand gets to measure about citations. One number, counting how often AI treats your own website as worth citing. The rest of the framework tells you how the answer came out, not where it came from.

The metric that should be there already has a name.

Citation source mix: of everything AI cited in the answer about you, how much came from pages you own, how much from coverage you earned, and how much from places you’ve never dealt with.

IAB’s own disclosure list asks vendors which kinds of sources they track and names editorial publishers among them, so the buckets exist. It just isn’t one of the metrics the framework asks for.

And that mix isn’t fixed. It moves with the engine and with the question.

Ask something early in a buying decision and earned sources dominate, but ask something transactional and your own pages do. Which is the argument for measuring it.

There’s a second gap and its shelf life.

The framework measures how fast a citation fades, but only for publishers. A news outlet can watch its citation of your story decay. You can’t, not for that coverage and not for your own pages. A citation is a position you re-earn, with no way to watch it slip.

Put those together and your PR and comms team can see what matters: the mix, which citations are slipping and what to go re-earn.

(Muck Rack publishes its nine source categories but not how any single link ends up in one, exactly the disclosure IAB would demand of a vendor. The Toronto paper ran 300 queries in one product category and sorted the sources with a language model.)

IAB’s framework says nothing about AI reasoning modes. It needs to.

AI platforms give people a choice most never notice: how hard should the model think before it answers.

There’s a fast mode for everyday questions and a high-reasoning mode for more complex work that takes longer, with settings in between. Same question, fast versus high-reasoning, and you get two nearly separate lists of citations.

I went through how ChatGPT decides what to cite in my last article. What matters for measurement is what happens to the source list.

Semrush and Kevin Indig of Growth Memo ran 100 prompts through GPT-5.2 on both settings. How often the answer cited a brand’s own site barely moved: 62.4% on fast, 60.6% on high-reasoning.

Underneath that flat number, Reddit and review sites lost about half their share, while government, academic, official documentation and support pages climbed.

Your presence holds steady. Everyone vouching for you gets swapped out.

If you count only whether you appear, you’d report nothing happened that day.

The frustrating part is that this one is easy to measure. Most AI visibility tools write their own prompts and send them to ChatGPT the same way you would, and whoever writes the prompt picks the mode.

IAB’s disclosure list already requires the platform, the model version, and whether the vendor let the model search the web first. But on how hard that model was told to think, it asks nothing.

What should a comms team do with this?

• Make a vendor answer the list. Ask which platforms and model versions they ran, whether the model got to search the web, which reasoning mode it was on and how they sort the sources they count. The last two aren’t in IAB’s framework, but they should be and you should ask anyway.

• Sort your own citation list. Ask the question your buyers ask before they know you, then put every source in the answer into three piles: your pages, coverage you earned and places you’ve never dealt with. Expect most of it not to be yours. That’s your citation source mix, and it costs an hour to find out.

Does it hold up?

One honest note: some of the studies I’ve cited in my own articles wouldn’t clear IAB’s bar. All the more reason why a common framework is needed.

Every company wants a way to measure AI visibility it can trust. IAB is where the ad industry settles what good measurement looks like, and we all know budget follows what gets measured.

Before your next budget meeting, ask whether a framework that leaves out one of the biggest parts of your GEO investment holds up.

That’s a tough case to make.

Subscribe now

1. Does ChatGPT search the web before it answers?

2. How does each source reach ChatGPT?

3. Which reasoning mode is ChatGPT using?

Change any one of them and the answer about your company changes with it.

You don’t get a vote on any of the three. You decide what ChatGPT has to work with.

What you’ll get in the next six minutes

• Whether ChatGPT goes looking

• The route a source arrives through

• Which reasoning mode it’s using

• What you’re left with: a citation you have to re-earn

• What I’d do tomorrow

Subscribe now

Does ChatGPT search the web before it answers?

ChatGPT doesn’t always search the internet to answer a question. Sometimes it does, but more often it answers from memory.

Semrush analyzed a 200 million user clickstream panel (a record of what real, opted-in users actually click on) and put the search rate at 34.5% as of February 2026, down from 46% in late 2024.

Two times out of three, nothing gets looked up.

When ChatGPT looks, you can still influence what it finds this quarter. When it doesn’t look, the answer comes from memory. Training data is the technical term, and it comes with a date stamped on it. The newest model’s memory stops on Feb. 16, 2026. Older ones still in service stop much earlier: GPT-5 on Sept. 30, 2024, GPT-4.1 on June 1, 2024, per OpenAI’s model documentation.

You cannot publish into the past.

So the company whose story has held together for years, across owned content, coverage, executive voices and the third-party record, has that consistency sitting in memory already.

The answer arrives on message with nothing retrieved. No other outcome in this system beats it, and only companies that built early get it.

A University of Toronto team measured how much of an answer comes from memory and how much from what ChatGPT looks up.

They found that a well-known brand’s position is mostly set in memory before anyone asks, which means a quarter’s worth of content probably won’t move it far. A challenger’s depends on what ChatGPT can find in the moment, and there’s usually not much there to find.

This cuts both ways.

Being the household name means the model has something to say about you. Now, whether that’s the story you’d have picked is another question, and plenty of large companies have never checked.

The advantage disappears the moment ChatGPT goes looking. Live search finds the clearest story, whatever the size behind it.

That’s pre-built reputation: the work is done before anyone asks.

(That study tested GPT-4o, not the models running today. It’s the best-documented work anyone has published on this and it’s a generation behind.)

How does each source reach ChatGPT?

When ChatGPT does go searching, the sources it finds don’t all arrive the same way. Some come in through ordinary web “search,” while others come through “news” or “Reddit” (ChatGPT’s own labels for the route a URL arrived on).

You don’t pick the route.

ChatGPT keeps track of how each one got there, and that matters more than the source itself.

Ahrefs went through 1.4 million prompts and found ChatGPT cites about half of what it pulls. Here’s the citation rate by the route it came in on:

• Ordinary web search: 88%

• News: 12%

• Reddit: 1.9%

• YouTube: 0.5%

• Academic: 0.4%

Ordinary web search is the only route cited at a high rate. News is next, and still 7x below it.

The frustrating part for PR teams: no one has published what the news route is.

We know the search route is ordinary web search, and we know Reddit has its own high-volume retrieval channel separate from search. Whether “news” is a filtered web search, a separate news index, or something attached to publisher agreements, I haven’t been able to determine.

So the 12% is the citation rate of a label that doesn’t have much public information behind it. A story you placed might arrive through the news route, through ordinary search, through both, or through neither.

If a vendor quotes you a number on earned media performance in AI answers, ask them about this.

(One note on this study. Ahrefs dates the data to February 2025, which can’t be right. GPT-5.2 wasn’t released until Dec. 11, 2025. February 2026 is the likely window. Either way, it isn’t the model running now.)

Does ChatGPT’s reasoning mode change which sources it cites?

The route a source arrives through explains part of ChatGPT’s citation rate.

The rest comes down to a mode in the interface that the person asking picks, usually without thinking about it, sometimes without knowing it’s there.

Semrush and Kevin Indig of Growth Memo ran 100 questions through GPT-5.2 twice on two different reasoning modes: once on Instant and once on Thinking.

Thinking mode searched the web 1,130 times. Instant mode searched 245. Thinking cited a source in 68% of answers against Instant’s 50%, and 4.5 sources per answer against 2.6.

Only 25.6% of the cited domains overlapped between the two modes. Same questions, same model.

On Instant, Reddit was 15% of citations and user review sites another 14.3%. On Thinking, Reddit fell to 7% and review sites to 6%, while official documentation and support pages climbed from 12.4% to 17.5% and government and academic sources went from 1.9% to 8.8%.

Ask on Instant mode and more of your story rests on Reddit threads and review sites.

Ask on Thinking mode and more of it rests on your own documentation and support pages, and on .gov, .edu and trade-association sources. Brands show up at about the same rate either way. What changes is who’s vouching for you.

Nobody picks this on your behalf.

Two people can ask the identical question about your company on the same afternoon and get answers built from two nearly separate sets of sources.

You don’t get to choose which room you’re in. You have to hold up in both.

(ChatGPT has since replaced that two-way choice with five levels: Instant, Medium, High, Extra High and Pro. The study’s two modes are the fast end and the high-reasoning end of the same dial.)

What is an AI citation worth?

That’s the three decisions. Say they all break your way and ChatGPT cites your page.

What did you get? Often a link with your name nowhere in the answer.

Semrush and Indig call these ghost citations. ChatGPT cites brands 87% of the time and mentions them by name in 20.7% of answers. Of the four engines they tested, that gap was the widest.

Then the citation drains.

Scrunch put the half-life of a ChatGPT citation at 3.4 weeks, measured across 3.5 million citations.

The pages themselves last far longer.

In the Ahrefs data, the median cited page in ChatGPT is around 500 days old. A page can be cited, dropped, and picked up again months later.

A citation is a position you re-earn.

(That’s an average across every kind of source, and nobody has repeated the study yet. Hold it loosely.)

What I’d do tomorrow

Three takeaways, and a first move for each.

• Build for memory and for live search. Ask ChatGPT your top category question, then ask it again with “search the web first.” The distance between those two answers is the distance between what it remembers about you and what it can find.

• Build for the fast answer and the high-reasoning one. Run the 5 questions your buyers actually ask on Instant and on a high reasoning mode, and save both source lists. Expect about a quarter of the domains to overlap. Those are two different assignments.

• Build the world that points at what you want to be cited for. Every citation decays, owned or earned. The page behind it lasts years, picked up, set down and picked up again. Write out the narrative you want your company known for, then go see whether you’ve built owned content on it for ChatGPT to cite.

Does it hold up?

You used to manage what people read about you. You now own the record that gets assembled when you’re not in the room, out of sources you didn’t choose.

Most of the time, without anything being looked up at all.

So go read your own record the way it gets read.

The documentation and support pages nobody has touched in a year. The Reddit threads and review sites that do more of the work in the fast answer. The last three placements.

Read them the way ChatGPT does, one at a time, cold, with no idea who you are.

Then ask the only question that matters: does it hold up?

Subscribe now

It’s The Crawl – Friday, July 31. Here are four things worth reading this week in public relations, communications and AI.

1️⃣ Answer first, elaborate later

New study shows nearly half of Google AI Mode’s citations — 47.7% — point at a specific paragraph rather than a page. On Pillarbase’s client pages, those passages ran a median 117 words, and 80% led with the answer.

The test is no longer only whether a page reads well, but whether a paragraph could be lifted out of it and answer a question.

Read more at Pillarbase.

2️⃣ Cited, but never named

Writesonic analyzed ~16 million brand appearances in a recent 30-day window. About 40% of AI citations didn’t name the source brand in the answer.

The range worth knowing: Perplexity 52%, Google AI Mode 49%, AI Overviews 41%, ChatGPT 37%, Gemini 25%, Grok 22% and Copilot 19%.

You influence the answer, but don’t show up in it.

Worth tracking whether your brand appears in the answer and not only in the source list, because those are different outcomes and a reader may only remember one.

Read more at Search Engine Land.

3️⃣ The brand holds, the citations churn

BrightEdge tracked e-commerce citations across ChatGPT, Gemini and AI Overviews for 12 weeks (May–July 2026).

The finding? The brands mentioned stay relatively stable, but the citations don’t.

On Gemini, only 3% of mentioned brands were new compared to the prior week, while 37% of cited URLs were new.

The name persists; the evidence behind it turns over.

Check this monthly and you will see a stable brand while missing that the pages behind it turned over completely, which is why a lost citation is not the same as lost ground.

Read more at BrightEdge.

4️⃣ GEO enters the PR mainstream

Muck Rack’s State of PR 2026 finds 73% of PR pros now call generative engine optimization at least somewhat important to their strategy. But 29% say no one at their organization owns it, and 39% aren’t measuring it at all.

The question has moved from whether GEO belongs in a communications plan to who owns it.

Read more at Muck Rack.

Subscribe now

Seven seconds later I had my answer, assembled from 46 sources.

The first citation was a LinkedIn post from the CEO. The second was another LinkedIn post from the CEO. Reuters came in third.

LinkedIn claimed the next 10 citations.

12 of the first 13 citations came from LinkedIn.

The CEO’s own posts outranked a global news organization. Twice.

What you’ll get in the next five minutes

• Three reasons Higgsfield AI should be on your radar

• The artifact: one question, 46 citations, a lot of LinkedIn

• The mechanic: what the CEO built on LinkedIn, and what backed it up

• What your comms team should steal

Subscribe now

Why you should know Higgsfield AI

Higgsfield is an AI video generation platform founded in 2023 by Alex Mashrabov, who previously ran generative AI at Snap. Higgsfield launched in early 2025. Keep that date in mind.

1. The Cannes film. In May, Higgsfield premiered Hell Grind at the Marché du Film at Cannes, the first fully AI-generated feature film made on its platform. 95 minutes, produced in two weeks, for about $500,000. The Wall Street Journal broke down the budget: $400,000 of it was compute. A comparable feature could cost around $50 million, according to Mashrabov.

2. The Fortune 500 footprint. 390 of the Fortune 500 use Higgsfield, and corporate customers drive more than 40% of its revenue. Both numbers come from the company’s first State of AI Video Report.

3. The valuation. In January, less than a year after launch, Higgsfield hit a $1.3 billion valuation with its latest funding round. Reuters wrote it up. So did TechCrunch and Higgsfield’s hometown San Francisco Business Journal.

Hold onto that last one. This is a company with real traction and real coverage from the exact outlets comms teams spend their year chasing.

The artifact: four words, 46 ChatGPT citations

I typed four words: “who founded higgsfield AI.”

ChatGPT’s answer:

“Higgsfield AI was founded in 2023 by Alex Mashrabov, its CEO, and Yerzat Dulat, its CTO. Mahi de Silva later joined and became the company’s third co-founder and Chief Strategy Officer.”

Sitting right next to the answer, a citation chip: LinkedIn +2.

Open the full source list and it gets more interesting.

Citation #1 is a LinkedIn post Mashrabov wrote about Yerzat Dulat, his CTO.

Citation #2 is a LinkedIn post Mashrabov wrote about Mahi de Silva, his third cofounder.

Two posts, written by the CEO, introducing the two people ChatGPT needed to answer my question.

Citation #3 is a Reuters article on the $1.3 billion valuation. Real reporting, ranked high. Two posts the CEO wrote about his own cofounders ranked higher.

After that, LinkedIn for the rest of the run, including:

• Mashrabov’s own LinkedIn profile.

• A video interview Mashrabov did, posted on AMD’s LinkedIn (a partner’s channel).

• Another interview, posted on TBPN’s LinkedIn (Silicon Valley’s SportsCenter).

• A post from Mashrabov linking to a sponsored interview he did on the podcast Silicon Valley Girl, which has 1.56 million YouTube subscribers.

• Three belonging to de Silva. Two posts are brand reshares with a line of his own added on top. The third is his profile page.

• A post from an industry voice marveling at what Higgsfield has built so quickly.

• A repost of a post from the Wall Street Journal’s Isabelle Bousquette, who shared video of a segment she did on BBC News talking about Higgsfield.

Follow the chain on that last one.

Bousquette wrote an actual WSJ article about Higgsfield and the Hell Grind premiere. She discussed it on a BBC News segment.

ChatGPT cited neither one.

It cited someone else’s LinkedIn repost of her post containing the video.

Still following?

The rest of the coverage landed near the bottom. Forbes on the company’s launch: #38. TechCrunch on the valuation: #42. The San Francisco Business Journal: #43. Out of 46.

Before I go further, let’s acknowledge the obvious. This is one search result that reflects a moment in time. It’s not scientific and I wouldn’t build a strategy on one platform.

But I would look at what it took for those two posts to be sitting at the top.

Subscribe now

The mechanic: how the CEO wrote the record himself on LinkedIn

I recently wrote about how executive thought leadership has a bigger job now. Every post, interview or keynote lands twice: once with its audience, and again as one more entry in the record AI reads.

Higgsfield is what that looks like in practice.

Start with the LinkedIn posts themselves.

Mashrabov wrote original posts introducing his cofounders, in his own words and on his own page. They told the market who built this company, and they published that answer where AI could find it.

Remember, when AI cites LinkedIn, 75% of the time it’s citing an individual, not a company page. The most-cited individuals: C-suite leaders.

Then there’s the coverage, the interviews and the rooms you might not think to count: a partner’s channel (AMD), an industry show (TBPN), a sponsored podcast (Silicon Valley Girl). Each appearance left another artifact on LinkedIn.

Finally, the outside voices, one of which reposted the WSJ reporter’s BBC segment.

A few things are playing out here:

• Higgsfield’s CEO putting original facts on the record: who built this company, what they did before and what the company has done since. None of it has to be pieced together from somewhere else.

• Earned media coverage and third-party validation backing those facts up. This gives those facts corroboration from sources AI already trusts.

• The information traveling, literally all the way into a LinkedIn repost, which correlates with higher AI visibility.

I asked about a company and the first two things ChatGPT reached for were things its founder wrote himself.

What your comms team should steal

• Build the credibility of your leadership team through your CEO. The top two citations for a founding question were posts where the CEO introduced his cofounders. ChatGPT’s answer to “who founded Higgsfield AI” was, in effect, written by the founder. When your CEO writes about a leader’s work in their own words, that post becomes part of the record AI retrieves when someone asks who runs the company. Write those posts on purpose. Then get the leaders themselves visible: three of the top 13 citations were from one of the cofounders, and two of those were company posts he reshared with a single line of his own on top.

• Plan the LinkedIn artifact alongside the appearance. The WSJ article mattered. The repost of the WSJ reporter’s segment is what got cited. Partner channels and podcasts. Every one of those rooms leaves one more citable trace. When you book the interview, plan where it lands on LinkedIn too.

The test takes 30 seconds.

Open ChatGPT and ask who founded your company. Read the first citation.

Does it hold up?

Subscribe now

Right now? The analyst who always gets quoted. The one your beat reporter calls and inevitably quotes saying something ridiculous, but whose framing still ends up in your earnings questions. (You just pictured yours.)

Or it’s the competitor CEO who’s willing to explain where the market is going and what companies like theirs are built to do about it.

When your execs aren’t building the public record, those voices are. And AI is drawing from that record to answer questions about your company and your category.

AI doesn’t leave a blank space where your authority should be. It fills that space with whatever holds together.

For years, executive thought leadership started with one question: What should this leader be known for?

AI search adds a second one: What ideas, evidence and language does this leader need to consistently put into the public record so that people and AI systems associate them and their company with a particular expertise or point of view?

The old job was placing your executive in the conversation.

Now that’s half the job.

The bigger half is positioning your executive’s ideas so they’re part of the answer.

What you’ll get in the next five minutes

• What the evidence shows (and what it doesn’t)

• The two stories AI search needs from your CEO

• How the leadership team opens more doors into the same world

• What it means for communications teams

Subscribe now

The evidence points one way: executive thought leadership can influence AI search

I’ve read every study I can find on how AI search picks its sources. Four are worth your time.

Princeton researchers added statistics, quotes and citations to content and boosted its visibility in AI answers by as much as 40%. Muck Rack traced where AI citations come from: 84% earned and third-party sources, with company-owned content adding another 13.7%. Ahrefs tested what predicts AI visibility and found branded mentions across independent websites correlated more strongly than backlinks or publishing volume.

Three studies, one pattern: AI favors ideas that arrive with receipts and keep traveling.

The fourth study is the one closest to home.

When AI cites LinkedIn, Meltwater found, 75% of the time it’s citing an individual, not a company page. The most-cited individuals: C-suite leaders. And 72% of the cited posts were original, written by the person posting them.

Now, none of these studies proves that your CEO can publish a few posts and magically influence answers across AI search.

What they suggest: thought leadership can influence AI answers when it does four things:

1. Introduces an original idea

2. Supports it with evidence

3. Earns validation beyond the channels your company owns

4. Comes from credible individuals, not just brand channels, and gets widely discussed

The executive post is one source. The interview is another. The byline, keynote, first-party study and resulting coverage add more.

Over time, they become a record that connects a leader to an idea and the idea to a company.

The CEO’s two stories: category narrative and company vision

If a CEO’s voice can become part of the answer, the next question is what they should consistently be teaching the market.

• The category narrative: what’s changing in the world around the company, why it matters and what the market needs to understand about what comes next.

• The company vision: why this company is built to lead or benefit from that change.

The category narrative defines the world. The company vision establishes the company’s role inside it. For AI search, each supplies a different part of the answer.

The company vision helps with questions like “Who is this company and what does it do?”

The category narrative creates the opportunity to show up in broader answers: “What’s happening in this space? Why is the market changing? Who has a credible point of view on it?”

The CEO’s job is to connect the two: build the frame others use to understand the category and the company’s place within it.

They need a credible answer to both. Without one, someone else (like that analyst) writes the frame.

Subscribe now

The leadership team: one story, many doors

One voice can’t make the whole case. That’s the leadership team’s role.

Their job is to deepen the corporate narrative through their own expertise. (Nobody needs five versions of the same spokesperson.)

The CEO frames what’s changing and why the company is built for it.

The CTO can explain the technical shift underneath that change.

The head of supply chain can show how the shift is reshaping the company’s distribution network.

The Chief Marketing Officer can translate it into customer and market implications.

Other operating and customer leaders can show what the strategy looks like in practice.

Each voice opens another door into the same world.

For people, that makes the company’s point of view more credible and useful. For AI systems, it creates a broader record connecting the company to specific areas of expertise.

Each leader adds authority the others can’t: a piece of evidence, a functional perspective, a customer example or a specific explanation of how the company is responding to the change the CEO has named.

They turn one big story into a body of specific, sourced stories.

And that body of work is what AI draws from when it assembles the answer.

What this means for communications teams

• Success has to mean more than reach. Placements, impressions and engagement tell you a piece moved. The stronger indicator is whether the pieces add up: a coherent record of executive content AI systems can find and connect to the company’s strategic narrative.

• Success is the body of work. Every post, interview or keynote lands twice: once with its audience, and again as one more entry in the record AI reads. Each should deepen a consistent point of view, add evidence or give another credible source something worth repeating.

• Start the program with the CEO’s two stories. Build the category narrative. Build the company vision. Then work with the leadership team to identify distinct ways each executive can substantiate both from their own seat.

The content calendar comes second.

The cost of waiting

A public record compounds as it accumulates.

Silence compounds too.

The longer someone else defines your category and who leads it, the more established their framing becomes. When your executives finally enter the conversation, they start against a record someone else has already built.

Building authority steadily is easier than trying to displace an answer the market (and the AI systems reading it) has already begun to repeat.

Does it hold up?

Executive thought leadership has a bigger job now: helping build the broader narrative AI systems can retrieve, verify and repeat.

So walk the program the way AI would.

Read what your executives have put into the world over the past year.

What change are they consistently helping the market understand?

What role does the company claim within it?

Do the leadership team’s voices deepen that story, or scatter it?

Then ask the only question that matters: Does it hold up?

Subscribe now

It’s Saturday (night), July 25. Time for this weekend’s Crawl.

1️⃣ If a citation disappears, does it make a sound?

OtterlyAI looked at 20 million URLs cited by AI over one month. 19.3% did not go to a live page.

Read more at OtterlyAI.

2️⃣ 20 years of homework, due now

Greg Jarboe says AI Overviews are increasingly citing YouTube. Most comms teams have built no video bench to be cited from.

Read more at Search Engine Journal.

3️⃣ The people want to know

Demand for AI visibility measurement is skyrocketing. Searches for “AI search tracking” are up 184% year over year. Searches for “AI rank tracking” up 175%.

Read more at Ahrefs.

4️⃣ So fresh, so citable

New Seer study finds 75% of pages LLMs cite were updated in the past year, but only 42% were published in it.

Read more at Seer Interactive.

I know because for the last year, I’ve consumed just about all of it, obsessively wrapping my head around what AI search means for comms, PR and my own team’s work.

Much of the advice is smart. Some of it I’d co-sign line by line. Nearly all of it is prescriptive.

Brands should prepare for AI search. Should build authority. Should invest in owned content. Should audit their digital footprint. Should, should, should.

The one thing I could never find was someone showing me what good looks like.

I wanted to see the receipt and learn from it: the actual page a brand built, the answer AI gives about them and the mechanic worth stealing.

The newsletter I wanted to read didn’t exist. So this is it.

What you’ll get in the next five minutes

• Why AI has raised the stakes

• What a narrative that holds up actually means

• What this newsletter will do about it every week

The stakes: AI search is the new room your company’s story gets told in

Reputation has always lived in rooms you’re not in. What the journalist says to their editor. What your employees say at dinner. Comms people have always known this; it’s why the work was never really about the press release.

There’s a new room now, and it’s the biggest one.

When someone asks an AI assistant about your company (is this brand credible? what happened with that spinoff?), the answer gets assembled in seconds from everything you’ve built and everything you haven’t.

Your owned content. The coverage. The third-party record. The stray forum thread. No call to your comms team. Presented as the answer.

This is the part that should reorganize how you think about the job: that answer can’t be spun in the moment. It reflects what exists, and nothing else.

If your narrative is coherent across everything you’ve built, that’s the story it tells. If there are gaps, they get filled with whatever else is out there, delivered in the same confident tone.

Reputation was already decided in rooms you’re not in. Now one of those rooms is open all night and never calls for comment.

What holds up means: world-building, literally

AI models and platforms will change over time. What I don’t believe will change is the premium they place on being accurate.

To get there, they look at what the world has written about you (in GEO terms, external signals): who said it, how credible it is, how new it is, whether others are saying it too. They weigh what they find and synthesize it into a narrative. Then they deliver that narrative to anyone who asks.

For companies, the implication is plain.

The story you want told about you has to be consistent and durable everywhere it’s told: news coverage, your own content, your executives’ bylines, social platforms. That’s the raw material the synthesis runs on.

A narrative that holds together survives the process and gets retold.

A narrative that doesn’t gets replaced, synthesized from whoever else has been building, competitors included. Or worse, you’re invisible.

Which makes communications teams something new: world builders. I mean that almost literally.

A good fictional world is an internally consistent place where many stories can happen and all of them feel true. You can enter through any door and end up somewhere coherent.

Corporate narrative works the same way. And AI just changed the physics of it: something now enters through every door at once. When the doors agree, you’re the answer. When they don’t, you’re a follow-up question.

That’s the bar this newsletter is named for.

A narrative that holds up is a world solid enough that every claim inside it survives contact with reality.

How you position the company. What you promise about the product. All of it tested by the systems that retell your story without you in the room.

Different rooms, same lesson

My career has been one long study in whether narratives hold: running external comms at a Fortune 500, three winning U.S. Senate campaigns, a governor’s office, a state’s chief elections official and a non-profit.

All of it playing out in national headlines and local markets across the country.

Different rooms, same lesson every time: the narrative that holds is the one built in peacetime. The teams that survived the bad day were the ones who’d spent years making the bad day survivable. The statement mattered, but the world it landed in mattered more.

What to expect every week

So, the practice instead of the prescription:

• The Walk: the longer piece, where I work an idea out. Some weeks it checks the doors of the world a company built, including what AI answers when asked. Some weeks it traces a mechanic, like how first-party data survives retrieval or how a citable narrative is built. Some weeks it follows one sharp idea. Different route each time; same question at the end: what holds up?

• Spotted: one real artifact from a brand doing it right, marked up. What they built, why it works, the mechanic you can steal.

• The Crawl: between issues, what’s worth knowing right now in the AI and PR space. Your reading list needs a filter.

Every example runs the same test: is it findable, is it citable, and is it consistent with the rest of the world the brand built?

You’ll see that test often enough to start running it yourself.

Last thing: I only show what good looks like. The bar for being featured is high: if it’s here, it held up.

The story that gets retold

The story you build today is the one that gets retold tomorrow, by people and now by the systems people ask, in rooms you’re not in.

Let’s build narratives that hold up.

Subscribe now