Silent Witness

“He said, she said” sounds like a description of a disagreement between two people. It is really a description of an information problem.

Two people were there. Something happened between them. Afterwards, their accounts differ. Nobody else saw enough to establish what happened, and the event itself has gone. What remains are two human recollections of something that can no longer be inspected directly. That is the purest form of he said, she said.

Add another person and the information problem changes immediately. It does not disappear: three witnesses can disagree, two people can lie and one can tell the truth, the majority can be wrong, and people can interpret the same behaviour differently. But there is at least another position from which the competing accounts can be judged. Add CCTV and it changes again. Add messages, photographs, location data, DNA, toxicology, injuries, timestamps, call records or people who saw the participants immediately before and afterwards, and it changes again.

That is already what we do.

When something serious is alleged to have happened between two people, investigators do not simply shrug and say there were only two people there. They look for everything around the event that might help reconstruct it. They examine the room, retrieve messages, look at CCTV, establish movements, take forensic samples, speak to people who saw the participants before or afterwards and ask what else survived.

They look for receipts.

Which is why he said, she said has always interested me in the context of the larger question behind this series: what happens when everything is recorded?

Not when more things are recorded.

Everything.

And if you are going to take that proposition seriously, he said, she said is an unusually clean place to test it.

The gap

Consider a first date between two strangers. Even today, a surprising amount of the evening may survive. The dating service may contain the conversation that led to the meeting. There may be a restaurant booking, CCTV showing them arriving, a card transaction for the meal. Their phones know something about where they went. Messages may establish what they intended to do afterwards. A taxi service may record the journey. Another camera may record them entering a building. There may be calls and messages later that night or the following morning.

If something serious is alleged afterwards, investigators can assemble those receipts. They may establish how much somebody had to drink, look for injuries, examine the room, talk to the friend who received a message afterwards and try to establish how the two people appeared when they arrived and when they left. None of that is remotely futuristic. The forensic system already attempts to reconstruct events from whatever survived them.

And yet there is often one enormous gap.

The door closes. Two people are alone. Later, one says one thing happened and the other says something else.

We may possess extraordinarily detailed receipts of the approach to that moment and equally detailed receipts of what followed, while the period at the centre of the dispute has disappeared. The forensic task is therefore to use what remains on either side to infer what happened inside the gap.

That is he said, she said.

When I first began thinking about a world in which everything was recorded, I followed the proposition into that room. What if the missing part wasn’t missing?

Not a consent button pressed at the beginning. Not a twenty-second clip in which both people say they want to be there. Not selected moments. If everything means everything, it means the encounter itself: the whole sequence, including whatever changes during it.

It is deliberately uncomfortable. It ought to be. If the proposition only works where recording is convenient, we haven’t really tested it.

Yet once you get past the discomfort, almost none of the underlying evidentiary machinery has to be invented. People already record intimate encounters. Police already encounter recordings relevant to crimes. Courts already deal with intensely private evidence. Investigators already construct timelines from physical and digital traces. We already understand that evidence can exist without being broadcast to the world.

My thought experiment changes only one variable.

The gap disappears.

I initially described that to myself as an increase in the amount of evidence: perhaps an investigation that would once have had twenty receipts eventually has two hundred, then two thousand. But that still misunderstands what completeness means. A three-hour continuous recording can be one receipt. If every relevant second survives, the number of receipts is beside the point.

The difference is continuity.

Today there might be a text message at 10.43pm, CCTV at 10.57, a taxi record at 11.06, another camera at 11.22 and a message the following morning. Investigators use those fixed points to constrain what could have happened between them. Every useful receipt makes the unknown space a little smaller.

Now remove the unknown space.

The conversation is there. The tone is there. The activity is there. The pauses are there. Changes of mind are there. What happened immediately before and afterwards is there. The forensic process has not acquired some magical new theory of human behaviour; it has simply been given much more of the thing it was previously trying to reconstruct.

So the interesting question is not how many receipts exist.

It is how much of the event is missing.

Silent witness

Suppose the whole encounter survives and, two years later, one participant accuses the other of sexual assault. What does the recording do?

It records whatever happened.

That sounds almost absurdly obvious, but its neutrality matters. If an assault took place, the record may establish it. If it did not, the same record may establish that. If somebody initially consented and later withdrew consent, the sequence survives. If somebody’s level of intoxication becomes relevant, their condition may be visible. Pressure, fear, aggression, enthusiasm, hesitation, misunderstanding and changes in behaviour may all form part of the same continuous receipt.

Perhaps one person’s recollection is substantially right. Perhaps the other’s is. Perhaps each remembers some parts accurately and other parts badly. Perhaps, once the receipt is examined, there is very little dispute about what physically happened but an enormous dispute about what that behaviour meant.

The receipt has no preferred answer. It has no allegiance to the victim or the accused, just as a fingerprint has no allegiance to the prosecution or defence. CCTV can place somebody at the scene or demonstrate that they were somewhere else. A message can substantiate an allegation or destroy it. Evidence is already offensive and defensive at the same time. Completeness merely makes that symmetry difficult to ignore.

The same record capable of sending somebody to prison could be the record that prevents somebody else from being convicted of something they did not do.

That is why Silent Witness feels right to me. The missing witness in he said, she said does not necessarily have to be another human being. It can be the surviving event. It does not remember badly, decide whom it likes or prefer the story that sounds more plausible. It still has to be interpreted, but it does not begin with a retrospective human account of what happened.

And that distinction matters because completeness does not eliminate interpretation. It changes where interpretation begins.

If there is no recording, we may first have to decide whether somebody said no. If the whole encounter survives and a refusal is clearly recorded, that factual question may disappear. Other questions remain. What happened after the refusal? Was it understood? Did the other person stop? Was somebody capable of consent? Was behaviour coercive? What does the law require in those circumstances?

Those may be extremely difficult questions, but notice where the disagreement has moved. We are no longer trying to reconstruct quite so much of a missing event. We are interpreting an event that survives.

Human interaction is temporal, which is why a complete receipt is almost the opposite of a simplistic consent token. A person can want something at one moment and change their mind later. Somebody can misunderstand. Enthusiasm can become discomfort. A twenty-second extract can look entirely different when the preceding hour is available. If everything means everything, sequence is part of everything too.

The strange result is that a continuous record could preserve more complexity than many of the accounts we produce afterwards. “She consented.” “She didn’t.” “He was aggressive.” “She seemed fine.” Those are compressions. The receipt preserves the material from which the compression was made.

It still does not tell us what to think. Receipts do not abolish interpretation. They give interpretation something harder to lie to itself about.

Living with it

There is then a different question: what happens to behaviour when people know that the gap will not exist?

Again, I don’t think we are starting from an entirely unknown human condition. We already live around cameras. Shops, streets, stations, offices, apartment buildings and homes are covered by CCTV. Doorbell cameras watch entrances. Cars record journeys and their surroundings. Nearly everybody carries a device capable of becoming a camera within seconds.

Most of the time we barely notice.

That does not mean observation is behaviourally neutral. Somebody deliberately making a video for an audience performs. A reality television contestant knows millions of people may be watching. Someone creating pornography for an audience is obviously behaving differently from somebody whose private behaviour is being recorded without their knowledge. Audience matters.

But persistent recording can also become infrastructure. CCTV is the obvious example. People still commit crimes in front of cameras, so observation plainly does not eliminate bad behaviour. Yet we continue to install cameras partly because we expect visibility to alter risk and partly because, when it does not, we want a record of what happened.

Think about something more ordinary. Why might somebody feel safer walking home along a well-lit street than along a dark one? Part of the answer is visibility. Other people may see what happens. There are potential witnesses. An act has less opportunity to disappear.

That is not a theory about some distant psychological future. We already organise environments around that intuition.

Now concentrate those conditions. Suppose both people on a first date know that if they later go somewhere private together, whatever happens will not simply vanish when the evening ends. Would someone meeting a stranger feel more confident because of that? Perhaps. Would someone inclined to push boundaries behave differently knowing that their actions could later be established rather than merely alleged? Quite possibly. Would both people pay more attention to changes in the situation because they know those changes will survive? Perhaps.

I don’t need to know the exact behavioural consequences to see the structural change. Human beings already normalise observation. Observation does not stop all bad behaviour, but visibility is already one of the things we use to alter risk. The thought experiment simply removes a space which has historically remained unusually resistant to that condition.

And there is no necessary reason for this witness to become an audience.

That distinction matters because sex is intensely private, and it is easy to hear “recorded” and immediately imagine “watched”. But we already know those are not the same thing. Evidence can exist without being published. Police can examine material that never enters the public domain. Courts routinely deal with information that remains highly restricted. A recording does not cease to be private merely because there are circumstances in which it can legitimately become evidence.

Perhaps future AI systems make this separation stronger. A machine might be able to interrogate hours of material for a specific question, locate the relevant passage and expose no more than necessary. But that is an implementation detail rather than the central idea. The important point already exists in the world we have: preservation and publicity are different things.

Privacy therefore does not necessarily require disappearance.

For most of human history, disappearance happened to be one of privacy’s most powerful mechanisms. Two people closed a door and, unless something left a trace, what happened behind it largely ceased to be available to anyone else. That protected intimacy, but the same absence also protected wrongdoing and made some false allegations impossible to disprove. The gap itself had no moral preference. It could protect either participant and endanger either participant.

A silent witness changes that condition without necessarily creating a public spectacle. The event can remain private while still remaining available in the exceptional circumstances in which establishing what happened becomes more important than keeping the record completely inaccessible.

We already make versions of that judgement today. Completeness simply makes the underlying trade-off harder to avoid.

When the gaps disappear

This future also looks less exotic when viewed through the devices and systems we already use. Increasingly, the usefulness of technology depends on its knowing more about us. A phone is useful partly because it knows where we are, who we communicate with and where we have been. Apps learn from behaviour in order to make recommendations. Watches monitor bodies. Cars sense their surroundings. Homes acquire microphones, cameras and connected appliances. We continuously trade information for navigation, personalisation, security, convenience, entertainment and health information, usually while insisting that the systems somehow preserve our privacy as well.

So the Receipts world does not require somebody to announce a universal surveillance programme and install a camera above every bed. It could arrive much more mundanely. Systems already surrounding us become more capable of remembering because remembering makes them more useful. The gaps shrink because the infrastructure becomes more continuous.

People growing up inside that environment may also have a different intuition about what should survive. I take very few photographs. My family never took many. Large parts of my own life exist only in memory, and that does not feel particularly strange to me because I grew up in a world in which most experience disappeared. Someone growing up now may already experience that differently. Someone growing up twenty years from now may find it extraordinary that so much of a person’s life once vanished without leaving anything searchable behind.

And even a complete recording of an intimate encounter would not really be the entire receipt, because events have edges only when we decide where to draw them.

Why did these two people meet? What had they said to each other during the previous week? What happened immediately beforehand? What had each person been drinking? What happened afterwards? Who did they contact? What did they say before anyone had a reason to think their words might later become evidence?

That is already how investigation works. The forensic question expands outward until the surrounding receipts make the central event intelligible. As more of ordinary life survives, that surrounding timeline becomes more continuous too.

Sexual encounters make the problem particularly stark because we care enormously about both privacy and proof, but the information structure is not confined to sex. Two executives leave a meeting with incompatible accounts of what was agreed. An employee says a manager threatened them; the manager denies it. A salesperson says they never made the promise a customer remembers. Two members of a family argue for years about a conversation that lasted five minutes. A politician says their remarks were taken out of context. A company says its system never targeted somebody in the way that person alleges.

Today we reconstruct. What emails exist? What messages survived? Who was there? What happened beforehand? What happened afterwards? What does the system log say? Where are the gaps?

The method does not change as the record becomes more complete. The amount of inference does.

For most of human history, the protection and the danger of a private encounter came partly from the same fact: nobody else knew exactly what happened.

That fact sits underneath he said, she said. The victim may have no way to prove what happened. The innocent accused may have no way to disprove what did not. The perpetrator may rely upon the absence of evidence, and a false accuser may rely upon precisely the same absence. Everyone else is left assembling whatever fragments survived and trying to reconstruct an event that no longer exists.

We already know what happens when some of those fragments survive. We call them evidence. When cameras capture part of an event, we look at the footage. When phones preserve messages, we read them. When forensic evidence contradicts someone’s account, we reconsider the account.

Nothing about the thought experiment requires a new theory of evidence.

Take the world we already have. Take the methods we already use. Take the way people already live around cameras, phones and observable environments.

Then change one variable.

Remove the gaps.

Not twenty receipts, two hundred or two thousand. The number was never the point. A single continuous receipt may be enough.

The event survives. Its sequence survives. What came before survives. What came afterwards survives. And one of the oldest information problems in human relationships begins to change because the condition that created it is disappearing.

We will still argue about meaning and motive. We will still misunderstand one another. People will still lie, and genuine evidence will still be capable of being interpreted badly. Completeness does not create omniscience, still less automatic justice.

But he said, she said has always depended upon an unwitnessed space between the two accounts.

What happens when that space is no longer empty?

There was a witness after all.

It just never said a word.

Timelines

On the evening of 10 September 2025, somebody ate dinner at Cowboy’s Smokehouse in Panguitch, Utah. The order went into the restaurant’s system at 8.55pm. At 9.47pm, $38.51 was paid on a Visa debit card. The till produced what tills produce: a record of an entirely ordinary transaction. A steak had been ordered, eaten and paid for. Nothing about the receipt itself knew that Charlie Kirk had been shot dead earlier that day at Utah Valley University, three hours’ drive to the north. Nothing about it knew that Tyler Robinson would soon be accused of the killing, or that months later people would be arguing about whether Robinson had been the man eating dinner in Panguitch that night.

Restaurant staff say they thought they recognised Robinson when photographs of him were released after the shooting and contacted the FBI. The owner subsequently told Fox News that investigators interviewed him and two servers, and that he supplied the last four digits of the card used for the meal. The restaurant’s payment system did not retain the cardholder’s name in the circumstances described, and the owner did not know whether investigators had subsequently established who owned the card. By January 2026, four months after the assassination, the possible sighting was public. In July, citizen journalist Diligent Denizen went to Panguitch, interviewed staff and showed the restaurant receipt itself on camera. One server said he was certain the customer had been Robinson. The owner, who had not been there that evening, was understandably more cautious.

The receipt does not prove who ate the steak. That is precisely what makes it useful. It proves something smaller: somebody ate there and somebody’s card was processed at 9.47pm. But place that mundane little object on a timeline and it begins to ask questions of other receipts. At Robinson’s preliminary hearing, the court heard that around 11pm his roommate, Lance Twiggs, received a message from Robinson’s phone. In the subsequent exchange, the sender said he was still in Orem because he was trying to retrieve the rifle prosecutors say had been left near the university. If the card used in Panguitch belonged to Robinson, those two things are difficult to reconcile. If it belonged to somebody else, the apparent contradiction begins to disappear.

We do not need to decide which by choosing whether to believe the server.

There should be another receipt.

Whose card was it?

That is the first thing the Panguitch receipt taught me. A receipt does not necessarily answer the question. Sometimes its most valuable function is to tell you what receipt you need next.

It also has a history of its own. It was created in September. Staff say its possible significance became apparent almost immediately. Information about it went into an institutional investigation. Its existence became public four months later. Months after that, the physical object itself entered a much wider public archive. Each of those moments is different. Creation, significance, acquisition and disclosure need not happen at the same time.

The receipt has a biography.

And once I saw that, Before Anyone Knew became considerably bigger.

Before Anyone Knew

I had been thinking about receipts for years, but almost entirely through the lens of marketing and MarTech.

Digital marketing records things because marketers want to know what happened. An advertisement was served. Somebody saw or clicked something. A customer visited a page, bought a product, abandoned a basket, opened an email, ignored another, came back three days later, bought something else. A system inferred something from that behaviour and changed what happened next. The entire industry spent decades building ever more detailed records because the ability to observe, measure and attribute behaviour was commercially valuable.

What interested me was the consequence nobody was really building the system for.

Those records create accountability.

If a company has recorded what it knew about somebody, what it inferred, what it showed them and what it did as a result, the same machinery that allowed the marketer to optimise a journey may eventually allow somebody else to reconstruct it. The company thinks it is building an attribution system. It may also be building the evidence with which its own decisions can later be examined.

That was the box in which I had put receipts.

The Charlie Kirk case took the idea out of the box.

I had not previously encountered the word being used quite so relentlessly in public argument. People following the case were not simply saying that somebody was wrong. They were saying: where are your receipts? And then somebody would produce one. A photograph. A message. A piece of video. A court document. A timestamp. A map. An archived webpage. A search graph. Another person would attack it, authenticate it, contextualise it or find another receipt which appeared to contradict it.

The argument had acquired objects.

Kirk was also an unusually revealing person for this to happen around because he was already one of the most receipted political figures of the platform era. His career had been built in front of cameras. Campus debates were filmed, clipped and redistributed. He hosted a daily show, had an audience in the millions and generated enormous volumes of commentary from admirers and opponents alike. He was deeply embedded in the political world around Donald Trump; after his death, the Washington Post described him as being on a first-name basis with an entire generation of White House officials.

Long before anybody knew which questions would eventually be asked about him, Kirk had already accumulated an extraordinary public archive.

Then the assassination generated another one.

More than 3,000 people were at Utah Valley University. Most had phones. The university had cameras. Other buildings had cameras. The FBI released imagery and appealed for more photographs, video and tips from the public. The difference from the assassination of John F. Kennedy is almost too neat. Abraham Zapruder happened to be in Dealey Plaza with an 8mm camera and produced one of the most examined pieces of film in history. Sixty-two years later, the problem surrounding Kirk was not a shortage of recording. It was abundance.

But abundance is not quite the point either.

The deeper change is that a receipt can exist before anybody knows the question that will later make it important.

That does not mean before the assassination. The event is not the boundary.

The Panguitch receipt was created afterwards. It still preceded the question which later gave it significance. A photograph taken six months after an event may acquire importance three years later because somebody notices who was in the background. A payment made next year might tell us something about a relationship which began this year. An old interview can lie unnoticed for twenty years until another object gives one sentence in it a completely different meaning.

What matters is not whether the receipt came before or after the event around which we eventually construct the story.

It is whether preservation came before significance.

That distinction seems small until you follow it through. For most of human history we preserved things largely because somebody already thought they mattered. A letter was kept because it came from an important person. A battle was recorded because a king wanted it remembered. An archive existed because somebody had decided which papers deserved space in it. Much of everything else simply disappeared.

Now vast quantities of ordinary life are preserved without anybody making that judgement first.

The till does not know which dinner will matter.

The camera does not know which pedestrian will matter.

The server log does not know which login will matter.

The payment system does not know which transaction will eventually become an alibi, a contradiction or nothing at all.

The receipt survives first.

The question arrives later.

Put It in Order

Once the question arrives, however, a warehouse full of receipts is not necessarily much use. Somebody has to make them legible.

One of the first things people do is almost embarrassingly primitive.

They put them in order.

That is one reason timelines have become so central to the continuing public examination of the Kirk case. Candace Owens does it. Lawyer and broadcaster Baron Coleman does it. Diligent Denizen does it. Other researchers synchronise footage, compare statements, locate documents and ask exactly when a particular object appeared.

Chronology can destroy an apparent contradiction simply by showing that two things happened months apart. It can create one by showing that an explanation came before the event supposedly responsible for it. A deletion means more when we know what happened immediately before it. A journey acquires significance when we know where somebody needed to be next. A statement by an institution means something different when placed beside what the institution had already received when it made the statement.

Panguitch is almost absurdly simple in that respect. Put 9.47pm in one place and around 11pm in another. The timeline itself proves neither story. It narrows the argument. It asks: who owned the card?

That is a very different way of thinking from treating a receipt as a gotcha.

The useful question is not always, What does this prove?

Sometimes it is, What should exist if this is true?

A card transaction touches a payment system. A piece of footage acquired by investigators should have an acquisition history. A tip sent through a formal channel may leave an intake record. A document accessed within a controlled system may generate logs. A file that disappears may have backups, metadata or traces of the deletion itself.

Absence is not proof. But once you understand the system well enough to know that something ordinarily leaves a receipt, an absence can become a coordinate.

This is where deletion becomes more interesting than simply making something disappear. Somebody who deletes an object is deliberately changing the record, but the act of deletion may create records they never intended to create. Someone who goes further and tries to hide the deletion is doing something else again. The attempt to interfere with the evidentiary chain is itself an event inside that chain.

And the opposite can happen too. People sometimes create receipts precisely because they anticipate a future question. Businesses keep audit logs. Tenants photograph damage before moving into a property. People document agreements because they know memories may later differ. At the darker end, somebody expecting suspicion may deliberately construct what they hope will become an alibi.

So receipts are not all innocent by-products of ordinary life. Some are deliberately preserved. Some are deliberately manufactured. Others are created inadvertently while somebody is trying to destroy or manipulate the record.

The investigator therefore has another question available: not simply what does this receipt say?, but why does this receipt exist?

You keep trying to reach the edge of the ledger and finding another ledger.

There is something else happening in the Kirk case which makes that recursion far more visible. Parts of the public have become extraordinarily knowledgeable about the receipts themselves.

That is different from merely having an interested audience.

Imagine the old model of a major criminal case. An expert gives evidence in court. A reporter hears it, writes an account and tells the public what was said. Most readers possess neither the underlying evidence nor enough contextual knowledge to challenge the description except at a very general level.

Now watch what happens when thousands of people have already spent months studying the same corpus.

They have watched the footage repeatedly. They know earlier statements. They have maps and documents open. They can search old podcasts. They remember that somebody described the same event differently six months ago. AI makes it increasingly easy to find the sentence buried inside a three-hour interview or compare testimony with hundreds of pages of filings.

So when somebody speaks in court, a portion of the audience is not hearing the evidence for the first time.

They are reconciling what is being said with receipts they already possess.

A witness can describe a sequence and somebody outside the courtroom can immediately say: hang on — what about this? An expert can characterise a piece of evidence while another expert elsewhere is looking at the underlying object. Even omission becomes more visible because you can only notice what has been left out when you already know what could have been said.

An omission still proves nothing.

But it can become a question.

That change matters because institutional authority has historically benefited from informational asymmetry. The institution knew the evidence. The public received an account of the evidence.

That asymmetry has not vanished. Investigators and courts possess powers and material the public does not have. But the public side of the equation has changed dramatically.

At sufficient scale, an audience can become a distributed analytical system.

A lawyer sees the procedural point. A pilot recognises something in aviation data. A local identifies a road. Someone familiar with video synchronises two cameras. A firearms specialist questions an interpretation; another specialist may then explain why that challenge is wrong. No individual possesses all the expertise, but an enormous interested population contains remarkable specialist diversity.

The public cannot issue warrants.

It can pay attention.

And attention is the thing almost every receipt needs before it can become significant.

Google Trends belongs here for exactly that reason. Coleman and others examining the case have published charts which they say show pre-assassination search interest in people and places that only later became publicly connected to the event — including Robinson, Twiggs, Timpanogos Regional Hospital and people associated with the medical response — with repeated apparent interest in Washington DC and Israel. That geography is one reason some people attach much larger intelligence-service theories to the material. But Google Trends does not identify individual searchers, much less intelligence officers; it reports sampled, anonymised, aggregated and normalised regional search interest, and Google explicitly warns that low-volume terms can produce statistical noise.

That does not make the material uninteresting.

It tells us what kind of receipt it is.

If one isolated low-volume spike appears, it may be noise. If researchers believe they are seeing a repeated pattern across multiple names which only later became significant, the pattern deserves to be tested rather than inflated into a conclusion. The provocative question is still there: why does the instrument appear to show interest, in those places, in things whose later significance was not yet public?

The Trends graph is not the answer.

It is another coordinate.

When the Receipt Appears

There is another property of receipts which becomes obvious once people start using them strategically.

Possession and presentation are different events.

Owens sometimes uses this almost theatrically. She makes a claim while already holding the object she believes supports it, allows critics to commit themselves publicly to a rebuttal, and then drops the receipt. The underlying object has not changed. What has changed is the timeline around its disclosure.

The person possessed it on Monday.

Everyone else discovered it on Thursday.

Those are different facts.

Courts institutionalise a much more formal version of the same distinction. The prosecution has evidence the public has not seen. The defence has material and arguments it may have no reason to present at a preliminary hearing. Discovery rules govern what must pass between the parties, but public presentation is another question again. Utah’s preliminary examination exists to decide whether probable cause is sufficient to send the charges to trial; it is not the trial, and neither side’s public presentation should be mistaken for a complete inventory of everything it possesses.

That makes the public record both extraordinarily rich and necessarily incomplete.

We know what has appeared.

We do not know everything that exists.

And increasingly we may know enough about the surrounding systems to know that certain categories of receipt should exist even if we have never seen them.

Panguitch gives us the cleanest possible example. We know a restaurant receipt exists. The owner says investigators received the last four digits of the payment card and interviewed staff. The FBI has not publicly said whether it matched Robinson. The answer may therefore have existed inside the institutional record for months while the public argument continued outside it.

That does not tell us what the answer is.

It tells us that the visible timeline and the possession timeline are not the same thing.

Once an institution acquires a receipt, something else begins. The receipt is received, logged, perhaps copied, compared, tested, ruled in, ruled out, passed to somebody else or eventually presented. Those actions are not merely administrative scenery around the evidence. They are events.

Events leave receipts.

That makes it possible to ask not only what happened in the original event, but what happened to the evidence afterwards.

When was it received?

Who had it?

What was done with it?

When did somebody else obtain access?

When was it disclosed publicly?

The question is not inherently accusatory. An investigation into a major crime will receive huge quantities of useless material. Most tips will go nowhere because most tips should go nowhere. Evidence can be excluded for perfectly legitimate reasons. Courts can restrict presentation. Investigators can examine a lead and eliminate it without holding a press conference announcing that they did so.

The important change is that the institutional process is increasingly capable of leaving a record too.

The investigation enters the timeline.

And once it does, the investigation can itself be investigated.

That is part of what has happened as Robinson’s case has moved into court. On 1 September 2026, Judge Tony Graf ruled that prosecutors had shown sufficient probable cause for Robinson to stand trial. Robinson has pleaded not guilty. The prosecution has presented video, DNA, communications and other material; the defence has challenged parts of that case, and the preliminary ruling expressly does not determine guilt or the ultimate weight of the evidence.

Yet the court is not entering a blank information environment. People outside it already possess footage, documents and months of accumulated argument. Some evidence introduced in court can therefore be compared almost instantly with an external archive. Conversely, the court and the parties possess evidence the external archive does not contain.

Neither ledger is complete.

And nobody owns the whole record.

That is a very different condition from the one in which the institution possessed the evidence, the broadcaster possessed the pictures and everybody else waited to be told what they meant.

A citizen journalist can travel to Panguitch and retrieve a receipt. A witness can speak directly to millions of people. A court filing can be downloaded and preserved by thousands of strangers. A deleted post can survive in screenshots. A three-hour podcast can become searchable. Someone can send an object directly to a creator whose audience has already spent months studying the case.

Those networks can produce nonsense with astonishing efficiency. A genuine receipt can be attached to a ridiculous interpretation. Coincidences can acquire enormous narrative weight. A person who has stared at the same footage for six months may simply become more confidently wrong about it.

But distributed custody has another consequence.

A receipt which once might have disappeared into a drawer can escape.

When Nobody Owns the Story

This is where more evidence produces a result which feels counter-intuitive.

It does not necessarily produce one story.

It can produce more stories capable of citing evidence.

The official case has receipts. People questioning parts of it have receipts. The defence will use receipts. Journalists and creators produce and acquire receipts. Then everybody argues about what those receipts mean.

The important word is mean.

A genuine receipt does not arrive carrying its interpretation.

The restaurant ticket is real and still does not tell us who ate the steak.

A genuine photograph can contain the wrong person.

A real timestamp can be attached to the wrong clock.

A video can be authentic while somebody misunderstands what is happening inside the frame.

A Google Trends graph can be an accurate receipt of what Trends displayed and still be a poor basis for the strongest conclusion someone wants to draw from it.

Receipts do not abolish interpretation.

They constrain it.

That may be enough.

A story that once depended entirely on authority can be forced to answer an object. A conspiracy theory can be destroyed by an object. Another theory can be narrowed because one piece survives while another does not. A witness can be supported by a transaction record. A witness can be contradicted by one. An institution can demonstrate that a supposedly ignored lead was examined. Or years later someone may discover that a receipt existed and ask why nobody seems to have done anything with it.

The argument keeps moving because every new receipt changes the shape of what can reasonably be said next.

That is why the language of receipts has become so useful.

It does not ask us to begin by deciding which storyteller deserves our trust.

It asks something narrower.

What exists?

What did it record?

When did it exist?

Who had it?

What does it actually establish?

And what receipt would we need next?

The Kirk case is unusually rich because all of those layers have accumulated in public almost at once. Kirk’s life was already densely recorded because he was famous. The assassination itself was recorded because cameras were everywhere. The hours afterwards generated transactions, messages, journeys and footage whose significance was not always obvious when they occurred. The investigation generated another record. The alternative investigation generated another. Court proceedings are generating another still. People are preserving the claims, rebuttals and receipts of everyone else.

And somewhere in Panguitch, in the middle of all that, a till printed a piece of paper.

At the moment of creation it was worth $38.51 to the restaurant’s accounts.

Then the future arrived.

Staff remembered a customer. Investigators received information. A journalist reported the story. A citizen investigator returned months later. The receipt itself became visible. People put it beside another point on the timeline. The card transaction became the next question.

Nothing on the piece of paper had changed.

Its significance had.

That is the thing the Kirk case made me see beyond the world of marketing.

We are not merely recording more important events. We are preserving more ordinary ones before anybody has decided whether they are important at all.

That changes the relationship between evidence and time.

For most receipts, nothing will ever happen. They will remain trivial forever. Nobody will care about the meal, the photograph, the login, the journey or the message. Almost every receipt is irrelevant almost all of the time.

But neither the person creating it nor the system recording it can know every question that might one day be asked.

And that means the future no longer always has to depend on somebody in the past having known what was worth keeping.

Preservation can precede significance.

The receipt can come first.

The question can come later.

The Prototype

Celebrity has always been an exercise in curation. Long before Instagram, studios, publicists, managers, photographers, newspaper editors and the celebrities themselves helped construct the person the public was allowed to see. Clothes were chosen, interviews arranged, photographs approved, stories placed and awkward stories discouraged. Sometimes names, biographies and whole identities were manufactured. Marilyn Monroe was not born Marilyn Monroe. The name, the photographs, the clothes, the poses and much of the public persona were part of the machinery that turned Norma Jeane into a star.

There was nothing particularly mysterious about this. Celebrity depended on attention, and attention was more valuable when it could be directed. Nor did that world depend on an absence of receipts. There were always receipts: a sketch made at an event, a letter, a diary, a newspaper report, a court record, something somebody saw and later described. Then came photography, radio, recorded interviews, film and television.

The difference was that distribution had bottlenecks. Most of those receipts reached the public through a relatively small number of routes: publishers, newspapers, magazines, studios, radio and television. If you could influence those routes, you had a reasonable chance of influencing which version of the person survived. Not complete control, but enough.

Celebrities also had something most people did not: people were paying attention to them. That distinction matters.

The Receipt You Didn’t Author

Paparazzi changed part of the bargain by creating a commercial market for receipts the celebrity had not chosen to create. In fact, the value of the photograph was often precisely that it had not been staged: the star leaving a nightclub, arguing with somebody, kissing somebody, looking drunk, looking ill, looking furious or simply looking ordinary. The celebrity was still curating, but somebody else was collecting, and somebody was willing to pay.

There was still another bottleneck. The photograph usually needed a newspaper or magazine willing to publish it. A recording needed a broadcaster or physical distribution. However intrusive the collection had become, distribution still required machinery.

The internet removed much of that constraint, which is why some of the early internet celebrity scandals felt so shocking at the time. Pamela Anderson and Tommy Lee are an obvious example. Their private recording was stolen in 1995. Intimate or compromising material involving famous people was hardly new. What was new was the environment into which this particular receipt escaped.

Once it was being distributed online, control of the original was no longer control of its evidentiary life. One copy could become ten, then ten thousand. Someone who had never met either person could possess it; someone else could copy it, save it, send it, mirror it or reproduce it again. Custody had become distributed.

That changed what denial meant. A celebrity could deny a story. A photograph made that harder, a recording harder again, and thousands of strangers possessing copies of the recording changed the problem completely. The public was learning what “viral” meant. Celebrities were learning what it meant to be the subject of it.

Delete It

Then came a smaller lesson which has become so ordinary that we barely notice how strange it is. A celebrity posts something. Perhaps it stays up for an hour, ten minutes or thirty seconds. They think better of it and delete it. Five years later somebody produces the screenshot.

People keep these things. Fans keep them, critics keep them, journalists keep them. Forums, old phones and archives keep them. Someone somewhere decided, for whatever reason, that this particular receipt might be worth retaining, and the person who originally produced it may have no idea who now has it.

This is where celebrities encountered the problem in Delete earlier than most of us. Deleting your copy does not delete everybody else’s, and deleting something is not the same as un-evidencing it.

Barbra Streisand provided the example that became a noun. In 2003 she tried to suppress an aerial photograph of her Malibu home. The attempt to stop people seeing the photograph instead caused vastly more people to seek it out. The attempt to suppress the receipt became another event, another receipt and another route through which the original travelled.

We eventually called it the Streisand effect. Celebrities didn’t need a name for the underlying problem. They simply had to learn to live with it.

What Celebrities Learnt

And this is the part I find most interesting. They adapted.

The old instinct did not disappear: deny the story, suppress the photograph, threaten the publisher, delete the post, persuade somebody not to run it. Sometimes those things still work. But once receipts are distributed, the calculation changes. If there is a receipt, assume somebody else may have it. If there are several, assume somebody may eventually put them together. If you describe an event one way, consider whether a photograph, message, recording or timestamp may later describe it another.

Denial becomes a more dangerous strategy when the receipt may arrive tomorrow, so curation changes. It doesn’t disappear; it moves downstream. Explain what happened. Put it in context. Apologise. Refuse to apologise. Say you have changed. Say you haven’t. Attack the interpretation rather than the evidence. Publish your account before somebody else publishes theirs.

The contest increasingly moves from whether the receipt exists to what the receipt means.

Receipts do not end argument. They constrain some arguments and make others necessary. Celebrities became unusually experienced at living inside that condition because they had no choice.

Some went further. They stopped allowing media owners to control all the distribution and collect all the money. The Kardashian family may represent one of the clearest mature versions of that adaptation. Whatever one thinks of them, they understood that if enormous attention was going to surround their lives anyway, there was value in owning more of the machinery through which that attention was created, interpreted and monetised.

The celebrity could now be the subject and the publisher, the broadcaster and the programme, the advertiser and the product. Social media accelerated that change because a famous person no longer necessarily needed a newspaper, magazine or television network to reach the people paying attention to them. The audience could be reached directly, and eventually sold to directly.

Something that had initially looked like a catastrophic loss of control became, for some celebrities, an extraordinarily valuable asset. They could not control all the receipts, but they could compete much more aggressively over their meaning, their distribution and the economic value created around them.

The Prototype

That is why celebrities belong in Receipts. They are not an exotic exception to the world I am describing. They are the prototype.

Fame historically created an unusual combination of conditions around an individual. More people watched them, more of what they did was recorded, more people had an incentive to preserve those records and more people had an incentive to find old ones. Most importantly, enough people cared for receipts which would have been utterly irrelevant in anybody else’s life to become significant.

The ordinary person had another kind of protection. Not privacy exactly, but obscurity. Nobody was looking very hard. The celebrity did not have the protection of the shoal.

That is why an interview from twenty years ago can suddenly matter, why an old photograph resurfaces, why somebody finds a forgotten post, and why a contradiction can be reconstructed by people who had no interest whatsoever when the original events occurred. The significance can arrive later. The receipt was already there.

Celebrities understand this unusually well. It is why public figures become wary of microphones they believe are switched off, why media training exists, why old posts are sometimes removed when someone becomes famous, and why lawyers, managers and publicists think about what may emerge as well as what has already emerged. They have learnt, imperfectly and sometimes painfully, to behave as though the receipt may one day be called.

And yet there is something else about celebrities that I find oddly reassuring. They didn’t stop living.

Despite knowing that they are watched more closely, recorded more often and that almost anything may resurface later, celebrities still behave much like everybody else. They go out, fall in love, argue, say stupid things, change their minds, make mistakes and occasionally do things they know they probably shouldn’t. Media training changes some behaviour at the edges, but it does not turn people into creatures constantly calculating the future evidentiary consequences of every ordinary act.

Perhaps more tellingly, none of this seems to have stopped people wanting to become celebrities. We have watched several generations discover the price of extraordinary scrutiny in public, yet fame remains something people actively pursue. In that sense, we may already have priced some of this in. Human beings appear perfectly capable of living with a much more recoverable life when there is sufficient value on the other side of the bargain.

That does not mean nothing changes. It means the biggest change may be in what can later be known about a life rather than in the ordinary business of living it.

Most receipts, after all, are irrelevant almost all of the time. The person who creates one cannot know every question that may later be asked of it. A joke nobody noticed in 2012 may matter in 2032. A photograph that appeared meaningless when it was taken may acquire significance because of something nobody knew then. An old interview may look completely different when another receipt appears alongside it.

The future brings its own questions. Celebrities have simply lived with more people asking them.

Removing the Fame Requirement

For the rest of us, that intensity of attention was historically unusual. Our lives always produced receipts too, but far fewer were captured, fewer still were distributed, and almost nobody had the time, money or inclination to assemble them.

Celebrity had already crossed each of those thresholds. Attention came first, then commercially valuable capture, cheap digital replication, distribution without custody and finally phones that placed a camera and recording device almost everywhere. Each step made celebrity life more recoverable.

Meanwhile the machinery that records ordinary life became commonplace too. Photographs, messages, purchases, locations, searches, work histories, interactions, likes, dislikes, relationships and responses all leave increasingly dense trails.

But one important difference remained between us and the celebrity: someone had to care enough to look.

Human attention was expensive. It made sense to spend hours reconstructing the history of a president, film star or footballer. It made very little economic sense to do the same thing for one ordinary person among millions. That was another version of the shoal.

Machines change that calculation. They do not need fame before they can pay attention. They can assemble, compare, retrieve and interrogate receipts at a scale that would have been economically absurd when the work had to be done by people.

That is why the history of celebrity suddenly looks less like a history of famous people and more like an early experiment. Celebrities were among the first people forced to discover that curation cannot completely control a distributed evidentiary record. They learnt that denial becomes harder when somebody else possesses the receipt, that deletion does not necessarily restore control, that an apparently private moment can acquire a public life they did not author, and that once the evidence escapes the remaining contest may be over context, interpretation and meaning.

Eventually some learnt to make money from the condition itself.

The rest of us are not becoming celebrities. Something stranger is happening.

Fame used to be one of the conditions that made a life extraordinarily recoverable. We are removing the fame requirement.

A Game of Whack-a-Mole

Imagine walking into somebody’s house in the middle of summer and finding a roaring fire in the grate. You would probably wonder what they had been burning. If you had arrived a few minutes earlier, perhaps you would have seen papers going into the flames. If you arrived later, you might find nothing but warm ashes. Either way, you could reasonably suspect that something had been deliberately destroyed.

But unless you already knew what had been in those papers, your knowledge might stop there. You could know that something had been burnt, perhaps roughly when, perhaps even who lit the fire, without having any idea what the papers said, how many there were, who wrote them, who received them or why somebody had wanted them destroyed. The destruction could therefore succeed even though the destruction itself was visible.

For most of human history, that distinction mattered enormously. Paper burns. Stone can be smashed. Names can be chiselled out of monuments. Letters can be thrown away. Files can be shredded. A witness dies. A private conversation survives only in the memories of the people who were there. Once the physical record disappeared, the information could disappear with it.

Destruction has never been perfect, of course. Someone else might have kept a copy. Another account might survive. An archaeologist might eventually uncover something that had been missed. Historians have always learnt to recognise suspicious gaps. But destruction could be very good indeed — good enough that the future might never know what it was missing.

That gave human beings an ability we have largely taken for granted: the ability to change what could subsequently be known by changing what survived. Kings did it. Governments did it. Companies did it. Criminals did it. Families did it. Lovers did it. Ordinary people did it every day without thinking of themselves as controlling history. They destroyed the record, and time did the rest.

That is becoming much harder.

The Pharaoh’s Archive

Now imagine an ancient ruler returning from a military campaign that had gone badly. The official account says it was a triumph. The monument records a victory. The court historian praises the ruler’s courage. Inconvenient reports disappear. The name of a disgraced general is removed. Letters from provincial officials are destroyed.

Centuries later, historians reconstruct the campaign from what survived. They may suspect that the official version is exaggerated. They may compare it with archaeology or with the records of neighbouring states. A contradictory inscription may turn up somewhere the ruler could not reach. But suppose he had been completely successful. The embarrassing documents are gone. Nobody recorded their destruction. Nobody outside his control retained a copy.

History inherits the version that survived.

Now change one thing. Imagine that we still do not possess the destroyed documents, but we have a record showing that twenty-seven documents concerning that campaign were destroyed on the ruler’s orders within three weeks of his return.

We still do not know what was in them. But we know something we did not know before, and more importantly, we now know where to look. Who wrote those twenty-seven documents? Who received them? Which officials were involved in the campaign? Which neighbouring kingdoms recorded the same events? What happened to the people associated with those reports? Did other records disappear at the same time? Were people promoted, dismissed or killed? Did expenditure change? Did the official language used to describe the campaign change afterwards?

The deletion has become information, and information directs attention.

The documents themselves may never be recovered, but their destruction has created coordinates for an investigation that otherwise might never have begun. That is a very different information environment from one in which successful destruction produces only absence, and it is much closer to the one we are creating.

What Does Delete Mean?

The word itself is wonderfully reassuring.

Delete.

We click a button and something disappears. The photograph leaves the gallery. The email leaves the inbox. The post is removed. The file is no longer in the folder. The message vanishes from the conversation. The interface tells us the thing has gone.

But what exactly has gone?

Usually, what we know with certainty is much narrower: we can no longer see the thing where we expected to see it. Perhaps the underlying data has genuinely been removed from the system. Perhaps it will disappear after a retention period. Perhaps it remains in a backup. Perhaps somebody else has a copy. Perhaps it was cached, indexed, forwarded or screenshotted. Perhaps another application received it. The details differ from system to system, and it would be silly to claim that no piece of digital information can ever genuinely be destroyed. Storage media can be physically destroyed. Systems can securely erase data. Records disappear every day.

The more interesting problem is certainty.

Once something has become digital, and particularly once it has interacted with connected systems or other people, how can you know that you have found every instance of it? You can delete every copy you control, but that is not the same thing as knowing that no copy survives. You can destroy the original device, but that is not the same thing as knowing that nothing the device did survives elsewhere. You can remove the webpage, but that is not the same thing as returning the world to the condition that existed before the webpage was published.

The moment information enters a connected environment, it begins acquiring relationships. Someone receives it. Something indexes it. Another system records access to it. A backup captures it. A search engine sees it. Someone quotes it. Someone takes a screenshot. Someone discusses it in another message. A database records that an action occurred.

Not every event generates every one of those traces. The point is that the original object is no longer necessarily the whole record, and that changes what deletion can accomplish.

A burning file can potentially destroy the information itself even while leaving evidence of the fire. A digital deletion may leave evidence of the deletion while the information survives somewhere else.

That is almost the opposite problem.

The World Around the Record

There is a temptation to think of a digital record as an object: a photograph, a message, a document, a post. But the object is only the easiest thing to see.

Consider a message between two people. There is the message itself, but there may also be a sender, recipient, time, account, device, network connection, notification, backup, reply, quotation or later reference to the conversation. Removing the message from one place does not remove all of those relationships.

The same is true of a webpage. The original page can disappear while links to it remain. A quotation can survive in somebody else’s article. A search result may show that it once existed. An archived copy may have been made. Someone may possess a screenshot. Another page may refer to what it said.

Sometimes none of those things happens. Sometimes all of them do. The important change is that the person trying to delete the thing cannot reliably know which world they are in.

That makes the phrase “escaped into circulation” increasingly difficult to define. Once upon a time there was a reasonably clear distinction between the private document in a locked drawer and the document somebody had published. Digital systems blur it. We create things on connected devices. They synchronise. They authenticate. They talk to other systems. Applications save state. Files move between devices. Services keep logs. Communications require information to travel somewhere.

The very point of much modern technology is sharing. We built an information environment whose fundamental advantage is that things can move effortlessly from one place to another, then discovered that movement makes recollection difficult to control.

This is why genuinely secret activity increasingly requires the deliberate manufacture of exceptional conditions.

Think of a spy operating in a foreign country for whom discovery would mean imprisonment or death. What device would you trust with the one fact that absolutely must never be discovered?

That question becomes uncomfortable very quickly. A supposedly secure phone is still a phone. A computer is still a computer. Both have histories, components, networks, software and surrounding systems. The extreme solution begins to look less technological, not more.

Leave the devices outside. Use a controlled room. Speak face to face. Perhaps write something on paper. Perhaps destroy the paper afterwards.

The important thing is not the paper. Paper leaves evidence too. Ink can have provenance. Paper can carry fingerprints or DNA. A document can be photographed. Another person can remember what it said. There has never been a magical medium in which information becomes impossible to discover.

The reversal is more interesting than that.

The safest way to delete a digital record may be never to create one.

Backwards Into the Future

There is something paradoxical about watching highly sophisticated people return to primitive methods of secrecy. Do not send the message. Do not take the phone into the room. Do not put the plan into the system. Do not photograph the drawing. Write it down. Memorise it. Destroy the physical object.

Technology becomes extraordinarily good at creating records and, at the extreme, the safest response is to avoid the technology.

The reason is not that a sheet of paper is inherently more secure than an encrypted device. It is that the paper can sometimes exist within a much smaller universe. One copy. Two people. One room. Destroy the copy and perhaps only two human memories remain.

A digital object can acquire witnesses merely by performing the function for which digital systems were designed.

The distinction is subtle but important. We used to create records and then protect them. Increasingly, if something genuinely must never become discoverable, the critical decision may have to happen before the record exists.

Do I allow this event into the information environment at all?

That is a much harder standard than good security, and it is a standard most ordinary life cannot meet. We do not live in secure rooms. We carry phones. We send messages. We make payments. We travel through cities full of cameras. We log into services. We communicate through systems designed to remember enough about what we did to function properly.

There is almost always a witness.

Not necessarily a person.

That is the difference.

Twenty-Seven Missing Documents

Return to the pharaoh and his twenty-seven destroyed reports.

Something else has changed since his time. An ancient historian who discovered evidence of those deletions would have to investigate them. Which people should be studied? Which archives? Which dates? Which neighbouring states? Which inscriptions? Which financial records? The absence may provide a clue, but turning the clue into an investigation is expensive.

Now put those twenty-seven missing documents into a modern information environment and give the investigator a machine that can search at a scale no human researcher could approach.

The machine does not need to recover the twenty-seven documents to make their deletion useful. It can ask what surrounded them. Which names appear repeatedly in surviving records from the same period? Which people communicated with the authors? What changed immediately afterwards? Which records elsewhere contain unusual gaps? Which documents refer indirectly to something that is no longer present? Which people altered their behaviour? What subsequent events would make more sense if some unknown event had occurred?

This is where artificial intelligence matters to deletion, but perhaps not in the way people usually imagine. AI does not make information indestructible. It makes residue more useful.

For a very long time, survival and discoverability were different things. A document could survive in an archive for a hundred years without anybody knowing it mattered. A reference could sit unnoticed in thousands of pages. A photograph could contain something nobody thought to examine. The sheer labour involved in looking protected the forgotten record almost as effectively as destruction.

Machines change that economics. Tell them what to look for and they can look in places no human investigator would have time to inspect.

And deletion itself helps tell them what to look for.

If twenty-seven ordinary documents disappeared at random, perhaps nothing follows. If twenty-seven documents concerning one particular campaign disappeared within days of an event, the pattern narrows the search. The deletion creates coordinates.

That does not make it proof. It makes it an investigative instruction.

This distinction will become increasingly important. A deleted message does not prove guilt. A removed webpage does not prove conspiracy. A missing photograph does not tell us why it was removed. A group of people cleaning up old social-media accounts may be doing nothing more interesting than cleaning up old social-media accounts.

But once a disappearance becomes visible, it changes the set of legitimate questions. What was there? When did it disappear? Who controlled it? Who else might possess it? What happened immediately before and afterwards? What surviving records touch the same people, places or times?

Deletion once closed investigative doors.

Increasingly, it may open them.

Editing History

This matters because the power to delete has never been distributed evenly. People with control over records have always possessed an advantage over people who merely appear in them. Governments controlled archives. Publishers controlled publication. Companies controlled corporate records. Courts decided which documents became public. Families controlled private papers. Victors controlled which version of events was commemorated.

The familiar phrase that history is written by the victors is incomplete. Victors have often possessed something even more valuable than the right to write. They had greater influence over what remained available to be read.

That does not mean the powerful always succeeded. Human beings have been hiding documents, leaking information, keeping copies and contradicting official histories for as long as there have been official histories. But successful suppression had a particular advantage: if done thoroughly enough, it could leave no argument behind. The future might never know that there had been another version.

Now imagine a historical record in which every meaningful alteration to the archive leaves some trace. Not necessarily the deleted contents. Just the fact of intervention.

These files existed. They were removed on this date. These people had access. These other records changed afterwards. Copies had previously been sent here. Someone searched for them immediately before they disappeared.

The archive no longer contains only history. It contains some history of what people tried to do to history.

That changes the balance. It does not eliminate secrecy or suppression. It does not guarantee truth. Power will continue to control information and people will continue to destroy evidence. But the old trick becomes harder.

Creating an absence is no longer necessarily enough, because the absence may have a shape.

And a shape can be investigated.

Delete Everything

There is a sentence people sometimes use when they are frightened of what might be found.

Delete everything.

In the physical world, under the right circumstances, that instruction could be surprisingly effective. Burn every copy. Shred the documents. Destroy the film. Remove the archive. If nobody else had the material and nobody knew exactly what had been destroyed, history could be returned surprisingly close to ignorance.

What would “delete everything” mean now?

It cannot simply mean every copy you possess. It would have to include every copy anyone else possesses, every backup, screenshot, indexed version, quotation and message discussing it. It might have to include the logs recording that it existed, the transactions generated because of it, the decisions influenced by it and the recollections of people who saw it.

And if you somehow had the resources to remove all of those, there is still another problem: what about the records showing that you removed them?

This is where the word everything becomes important.

It is easy to imagine a world in which everything is recorded and think about cameras, microphones, messages and databases. Those are the obvious receipts. But if everything means everything, it cannot stop at the original event.

Something happens and leaves a receipt. Someone copies it and leaves another. Someone accesses it and leaves another. Someone changes it. Someone deletes it. Someone notices the deletion. Someone investigates why it was deleted. The attempt to erase the record becomes part of the record.

You keep trying to reach the edge of the ledger and finding another ledger.

This does not mean nothing can ever be made secret. It does not mean no information will ever disappear. It certainly does not mean every human action will one day be reconstructed perfectly.

The claim is simpler and, I think, much more consequential.

Once something has properly entered the information environment we are building, returning the world to the condition in which nobody can ever discover that it happened becomes extraordinarily difficult.

Deleting the file is one thing. Un-evidencing the event — removing it from future knowledge — is another.

For most of history, time helped us. Memories faded. Objects decayed. Papers disappeared. Witnesses died. Archives burnt. Ordinary events left little behind. Sometimes that protected criminals and tyrants. Sometimes it protected ordinary people from being defined forever by their worst day. Sometimes it simply meant that almost everything human beings did eventually became unknowable.

We are changing that condition without having fully noticed what it was doing for us.

The future may increasingly be able to ask questions of the past that nobody living through the past expected anyone to ask. And the evidence may not be the thing somebody tried to preserve. It may be the thing somebody tried to destroy.

The twenty-seven missing documents may matter precisely because they are missing. The altered page may matter because we can see that it changed. The erased message may direct us towards the person who received it. The hole may tell the machine where to dig.

That is what makes Delete such an odd word for this new world. Deletion used to be an attempt to end the informational life of something. Increasingly, deletion may become another chapter in that life.

The fire can still burn the papers.

But the warm fireplace remains.

And now, more often than before, the future may know where to look for another copy.

Safety in Numbers

In the late 1990s, when people first started worrying seriously about privacy on the internet, I had an answer. I told them they were protected by the shoal.

I may occasionally have called it the herd. I cannot remember. It was nearly thirty years ago and I was not trying to develop a theory of privacy. Most of the time I was trying to persuade somebody that putting their credit-card details into a website was not an act of madness. But shoal is the better word for what I meant.

Imagine millions of fish moving through an ocean. Every one of them can be seen. Every one leaves some disturbance in the water. If you know exactly which fish you want, and you have sufficient reason to pursue it, you may be able to follow it. But why would you? There are millions of them. That was more or less my argument.

It is important not to rewrite my younger self as some sort of privacy visionary. I wasn’t. I worked for Excite, then one of the major internet search engines, and I had a very obvious commercial interest in the internet succeeding. I wanted people online. I wanted advertisers online. I wanted commerce online. I thought the internet was going to be enormous.

At a time when many perfectly intelligent people were frightened of typing a credit-card number into a web browser, I was often on the other side of the argument saying: look at what this makes possible. I sometimes regarded the emerging privacy debate as badly thought through, perhaps even as a restraint on trade. That position makes more sense if you know where I had come from, because I did not begin my career in technology. I began by selling audiences.

Selling Audiences

My first media jobs were in specialist publishing. A technical magazine might have 50,000 subscribers, perhaps people working in information technology or another tightly defined profession. Advertisers did not really buy the magazine; they bought access to those 50,000 people. The magazine was the interface. The content was what persuaded the right people to read it. The list told the advertiser who those people were.

Later I worked in national newspapers, including The Times, where the audiences were much larger but the underlying economics were remarkably similar. Reach and frequency mattered. If your publication reached 1,000,000 people and mine reached 400,000, I had a problem. I could argue that my audience was better: perhaps my 400,000 readers were exactly the people the advertiser wanted, richer, more senior, more interested or more likely to buy. Sometimes that was true. But “quality” was also very often the argument available to the person who had less quantity. If your 1,000,000 contained my valuable 400,000 anyway, plus another 600,000 people who might also buy the product, your scale was difficult to beat.

Media people understood this instinctively. The person able to create the greatest useful exposure was in a powerful position. The great weakness of that model was what happened next.

You could deliver the advertisement, estimate the readership, calculate opportunities to see, run research asking people what they remembered and examine sales before and after a campaign. But between the advertisement and the purchase sat an enormous area of ambiguity. Someone sees an advertisement for a car on Tuesday, another on Thursday, a friend mentions the car on Saturday, they drive past a dealership on Sunday and three months later they buy one. Which thing worked?

The old joke, attributed in various forms to generations of advertisers, was that half the advertising budget worked; the problem was knowing which half. The joke survived because the ambiguity was real.

The Click

Then the internet arrived. What interested me most was not that a magazine page had become a webpage. It was that the loop could close. A person could see an advertisement and do something immediately. No coupon to cut out, no telephone number to remember, no salesman asking, “And where did you hear about us?” They could click. The response existed inside the medium itself.

Early internet advertising nevertheless looked remarkably like the media business it was replacing. Advertisers still bought exposure. Banners were sold on impressions. The terminology changed — an “opportunity to see” became an impression — but the economic idea was recognisable. Almost immediately, people started distrusting it. They ignored banner advertising. They questioned whether an impression meant anybody had really seen anything. They wanted to know why they should pay simply because an advertisement had appeared on a screen.

Clicks looked much more attractive. They appeared to solve the problem advertising had always had. Instead of paying for the possibility that someone might notice the advertisement, why not pay when somebody actually responded?

I was an early enthusiast. I sold one of the first click-based search deals in the UK, and I could see the logic immediately. Search made the proposition particularly powerful because the person had declared some sort of intent and the advertiser could pay for the response rather than simply the exposure. Later, when Google built its business around search advertising, the power of the model became obvious to all.

The industry then spent the next quarter-century going further and further down that road. Clicks became engagement, engagement became conversion, and then came swipes, views, follows, purchases and every other measurable action that could move us closer to an outcome and further away from the ambiguity of the impression.

I have come to think we made a considerable intellectual mistake along the way. We started treating the things we could easily measure as though they were the only things that mattered. Exposure never stopped mattering simply because clicks became easier to count. But that is another argument. For this story, the important development is that the audience was becoming visible at higher and higher resolution.

If I know that you read a particular newspaper, I know something about you. If I know that you read the business section every morning, I know more. If I know that at 8.17 this morning you searched for flights to New York, I know something different again. If I know which result you clicked, more. If I know that you came back tomorrow and searched for hotels, more again.

The internet did not invent targeting. Media had always segmented audiences. What the internet did was steadily narrow the distance between an audience and a person.

And yet I still was not particularly worried - because of the shoal.

I knew the data existed. That was not in dispute. I also knew that if somebody had sufficient reason to investigate one individual, the protection became much weaker. If you were suspected of terrorism and an intelligence service decided that you mattered, that was different. If the police were investigating you, different. If you were enormously wealthy, politically important, famous, dangerous or otherwise interesting enough to justify the effort, different.

My argument was about ordinary people. Why would anybody expend the resources necessary to isolate one entirely unremarkable person from an ocean of information? The more data the internet produced, the stronger that argument appeared to become. The shoal became bigger. You were not invisible. You were obscure.

By the end of the 1990s, the advertising industry was already developing ways of making that ocean easier to navigate. In 2000, the US Federal Trade Commission described advertising networks using cookies to track behaviour across otherwise unrelated websites and build profiles attached to particular browsers or identifiers. In some circumstances those behavioural profiles could then be connected with identifiable information and offline purchasing data.

Even then, however, there was an economic problem. The dream was the individual. The practical unit was usually the group.

Peppers and Rogers

The ambition to treat customers one by one was certainly not invented by artificial intelligence. Don Peppers and Martha Rogers had popularised the idea of “one-to-one marketing” years earlier. In 1999, writing with Bob Dorf in Harvard Business Review, they described the concept as changing a company’s behaviour towards an individual customer according to what that customer had told the company and what else the company knew about them.

But the same article contained the problem. The mechanics were difficult. Identifying an individual, tracking them, interacting with them and then changing the organisation’s behaviour towards that particular person was much harder than telling a salesperson to pay attention to the customer standing in front of them. The aspiration was one-to-one. The economics forced us towards something else.

That something else was the bait ball.

A bait ball is what happens when predators compress a shoal of small fish into a dense formation. The individual fish is much easier to consume as part of a concentrated group than it is swimming freely through the ocean. Marketing has always created its own versions: readers of this magazine, viewers of that programme, people in this postcode, women aged 25 to 44, households with this income.

The internet made the bait balls both bigger and more precise. People searching for mortgages. People visiting car websites. People who had bought running shoes. People who looked like people who had bought running shoes. People who had visited a page but had not bought. People who had bought six months ago but not since.

The industry calls this segmentation, and segmentation is normally described as evidence of sophistication. Customers are different, therefore a sophisticated marketer divides them into useful groups and treats those groups differently. That is true. But there is another way of looking at it.

We grouped people because treating everybody independently was too expensive.

The bait ball was a compromise.

Then we did something rather extraordinary. We persuaded the fish to help construct it.

Search for something, click something, watch something, stop watching it, like, dislike, follow, unfollow, join, leave, share, ignore, buy, return. Every one of those behaviours performs an obvious function for the person doing it, but it does something else at the same time. It produces information about the person.

The modern internet is full of these dual-purpose actions. You experience a piece of content; the system experiences another observation. YouTube says its recommendation system learns every day from more than 80 billion signals, including watch history, search history, subscriptions, likes, dislikes and explicit indications that somebody is not interested. Its systems use those signals to predict what an individual viewer is likely to want next.

Put that back into the language of the shoal and something extraordinary has happened. The fish are continuously signalling, and the water is continuously listening.

User-generated content accelerated the process enormously. People gathered themselves around almost every conceivable interest: football, fashion, parenting, politics, diet, religion, cars, finance, conspiracy theories, comedy, sex, knitting, celebrity, disease, technology, grief. The platform did not have to employ editors to invent every category into which humanity might usefully be divided. We did much of it ourselves.

Creators gathered audiences. Audiences revealed preferences. Preferences generated behaviour. Behaviour improved the ability to predict what each person might respond to next. The shoal increasingly participated in its own profiling.

The Wrong Word

For most of this period, I continued to think about privacy in broadly the same way I had in the 1990s. Yes, technology knew more. Yes, profiles became richer. Yes, targeting became more sophisticated. But there were still limits to how much attention could actually be devoted to one ordinary person.

Then I realised I had been using the wrong word.

The shoal had protected us. Privacy hadn’t.

Privacy has never protected us.

What protected us were practical constraints that we mistook for privacy: too much information, too little storage, too little processing power, too many databases that could not easily talk to one another, too much labour involved in connecting the traces, too much effort required to work out what they meant, too few people available to watch everybody and, above all, too little economic reason to devote sustained attention to an ordinary individual.

That bundle of constraints produced something that felt like privacy. Most of us went through life largely unexamined not because examining us was impossible, but because examining everybody was impractical.

The shoal did not hide the individual fish. It made the individual fish expensive.

That is what has changed.

One by one, almost every component of that protection has become cheaper: recording, storage, transmission, search, correlation, prediction. The volume of data has continued to explode, but the cost of finding something inside it has fallen. What once required a person sitting in front of several databases can increasingly be done automatically. What once required an analyst to notice a pattern can increasingly be inferred by software.

Attention

And now artificial intelligence is attacking the constraint I think matters most: attention.

Human attention is expensive. That fact has shaped almost every institution we have built. A salesperson can get to know twenty important customers extremely well, perhaps fifty, perhaps a few hundred with help. They can remember that one customer’s daughter has started university, another hates telephone calls, another normally buys in March, another responds badly to pressure. That knowledge can make the salesperson extraordinarily persuasive.

But the relationship does not scale indefinitely because the human being doesn’t.

The same has always been true of politics, fundraising, intelligence work, fraud, propaganda, friendship and manipulation. The closer the attention, the fewer people could receive it. That was one of the great invisible constraints on individual influence.

AI changes the arithmetic.

Recommendation systems have already shown what happens when machines can make slightly different decisions for millions or billions of users simultaneously. Meta says it personalises experiences for billions of people using its services each day. Generative AI adds another capability: it can talk back. It can alter the language, respond to an objection, remember previous interactions and try something different next time.

Researchers are already beginning to measure what happens when those capabilities are combined with information about the person being addressed. A 2024 series of experiments involving 1,788 participants found that ChatGPT-generated messages tailored to psychological characteristics were more influential than equivalent non-personalised messages across consumer and political contexts. The important point was not that machines had discovered persuasion. It was that generative AI could automate and therefore scale personalised persuasion that previously required much more human effort.

That distinction is the whole story.

We have known for thousands of years that the best way to persuade two different people may be to speak to them differently. We have known for decades that businesses would like one-to-one relationships with their customers. We have known since the beginning of digital advertising that more information allows more precise targeting. None of that is new.

What is new is the economics.

The person has always been the ideal unit of attention. The audience was necessary because the person was too expensive. The segment was necessary because one-to-one did not scale. The bait ball existed because pursuing every fish separately consumed too much energy.

Now imagine that constraint disappearing.

Ten million people can still end up in the same place. They might buy the same product, support the same candidate, adopt the same belief or join the same movement. But it no longer follows that those ten million people have to receive the same intervention to get there.

One person might need reassurance, another excitement, another fear, another repetition, another social proof, another discount, another argument, another image, another twenty encounters over six months. If the system can observe the response to each intervention, remember it and modify the next one, the group can become the result of individual treatment rather than the unit through which the treatment was delivered.

Which brings me to a word the marketing industry will not like: Grooming.

We prefer engagement, personalisation or relevance. Those words are not necessarily wrong, but they describe the process from the point of view of the organisation doing it. Grooming exposes something the softer language hides. Grooming is longitudinal. You observe, learn, intervene, see what happened, remember, adapt and intervene again.

It does not require that the objective is sinister. The mechanism is the important thing. A brilliant salesperson does it. A political campaigner does it. A confidence trickster does it. A teacher can do it. A friend can do it. An abusive partner can do it. The morality comes from what the person is being moved towards, the degree of agency they retain and who benefits. But the mechanism is remarkably similar.

Repeated, personalised, adaptive attention changes people.

Historically, that kind of attention was scarce. That scarcity protected us — not perfectly, not equally and not by design, but materially. A government could intensely scrutinise some people; it could not afford to intensely scrutinise everyone. An advertiser could study its best customers; it could not maintain a unique, responsive relationship with every potential customer on earth. A political campaign could knock on doors; it could not give every voter a permanent political companion who remembered every conversation. A manipulator could groom one person; they could not personally groom fifty million.

Those were not ethical safeguards. They were physical and economic ones.

And we built a civilisation in their presence.

That is what I had not understood when I reassured people about the shoal. I thought the existence of all that data was itself part of our protection. In an important sense, I was right. The data was the shoal. There was so much information and so little capacity to pay attention to each individual trace that most of us were not worth isolating.

Privacy was what we called the condition. But privacy was not what produced it. Friction did. Cost did. Scarcity did. Human limitation did.

Over the last thirty years we have systematically dismantled those limitations. We recorded more, stored more, connected more, searched more, measured more and profiled more. While all of this was happening, we spent our lives producing the receipts that made each of us progressively more legible: what we watched, what we bought, what we ignored, what we returned to, what made us angry, what made us stay and what made us move.

For a while, the sheer quantity of those receipts still helped protect us. There were simply too many to read.

That may be the protection AI finally removes.

The shoal is still there. It is larger than anything I could have imagined in the 1990s. The difference is that the machine can now begin to pay attention to every fish.

Privacy has never protected us. What protected us is being stripped away.

Do you agree?

Most Email A/B Tests Can’t Deliver What We Pretend They Do

By now, the pattern should be unmistakeable.

Part 1: Sample Size: showed that most lists can’t detect realistic lifts without absurd sample sizes — so tests are often underpowered by design

Part 2: The 24-Hour Mirage: showed that even when you do see a “winner”, the measurement window is usually too short for the outcome you actually care about, and the testing workflow often changes the send conditions.

Part 3 The Missing Control Group: showed that without a baseline, you can’t separate your “learning” from the week itself.

Put those three together and you get a conclusion that’s hard to avoid:

Most email A/B testing is structurally incapable of producing reliable knowledge.

Not because marketers are incompetent. Because the constraints don’t match the claim.

What marketers want from testing (and what testing can actually provide)

Most teams treat A/B tests as if they’re doing science. They want three things:

1. a clear winner

2. a result that will hold when repeated

3. confidence that the change caused the outcome

Those are reasonable desires.

But in email, at typical list sizes, with typical measurement windows, and typical operational constraints, you almost never get all three.

What you usually get is this:

• a winner that’s “conclusive enough”

• within a window that’s convenient

• on a sample that can’t prove small lifts

• inside a system where baseline drift is invisible

That is a mechanism for making decisions, not for producing truth.

The real contradiction: email isn’t a lab environment

Email marketers behave as if they’re running lab experiments.

But email isn’t a lab channel. It’s a live publishing system.

You’re not testing in a sealed environment. You’re testing in the open world:

• your audience composition is drifting constantly

• your deliverability environment changes

• your competitive environment changes

• your product environment changes

• your news and seasonality environment changes

So even if you could solve statistics and timing, you still face the hardest problem:

the world won’t hold still while you run your test.

Why this forces “safe tests” — and why safe tests rarely teach anything

Here’s the practical organisational truth.

A true test requires the willingness to be wrong at scale.

If you want enough sample to detect meaningful lifts, you have to expose a large portion of the list to a truly uncertain variant. That means accepting the possibility of a real hit.

Most teams won’t do that. They can’t.

So they do the rational thing: they test variants they already believe are safe. Minor differences. Cosmetic changes. Familiar offers.

But the more “safe” your variants are, the smaller the true effect size is likely to be.

Which means you are now trying to detect tiny differences with tiny samples…

…in a moving environment…

…inside a short window…

…without a baseline.

That’s not “bad testing”.

That’s structurally non-informative testing.

The ESP problem (without blaming ESPs)

This is also why platform A/B tools are built the way they are.

They’re not designed to run lab-grade experiments. They’re designed to produce an answer that a normal team can act on this week.

They are decision systems. They are not truth systems.

That’s not a criticism of the platforms. It’s a description of the product requirement: most customers do not have the list size, patience, or risk tolerance to do true experimental design.

So the platforms optimise for closure.

What this means (and what it doesn’t)

It does not mean:

• “never test”

• “your instinct is better than data”

• “email is unknowable”

It means something more useful:

Stop pretending that one small A/B test gives you durable knowledge.

Most A/B tests, as commonly run, can only responsibly be treated as:

• directional

• local

• time-bound

• and subordinate to programme-level measurement

In other words: they can help you avoid obvious mistakes and occasionally discover a large effect.

They are not a reliable engine for incremental truth.

The simplest honest rule

If you want to keep A/B testing without lying to yourself, you need one rule:

Only claim what your test design can support.

If your sample is small, your claim must be small: “directionally, this looked better in this window.”

If your window is short, don’t declare a long-term winner.

If you don’t have a baseline, don’t over-credit the creative.

And if you ran a test designed to be conclusive, don’t mistake it for being correct.

The clean takeaway

Email A/B testing has become a ritual because it offers something organisations crave: a story of control.

We tested. We learned. We optimised.

But the uncomfortable truth is that most of the time, what you’re calling “learning” is simply a decision you would have made anyway, with a dashboard attached.

If you want better outcomes, you don’t need more tests.

You need more honesty about what tests can do — and a measurement system that treats email as what it actually is:

a long-running programme, published into a moving world.

Catch up:

Part 1: Sample Size, Part 2: The 24-Hour Mirage, Part 3 The Missing Control Group

I write about the system‑level mechanics behind email, attention, and digital behaviour.

The Don’t forget to subscribe 👇

Most email A/B testing conversations assume the same thing: if performance changes, it must be because you changed something.

Subject line. Offer. Creative. Send time. Segment. Frequency.

But in real programmes, performance shifts for reasons that have nothing to do with your experiment.

The market…

Read more

Why “Picking a Winner” Fast Usually Selects for the Wrong Thing

Email A/B testing has a hidden bias that almost nobody talks about.

It isn’t statistical.

It’s temporal.

Most email tests are judged inside a measurement window that fits a workday. Sometimes a few hours. Sometimes “by tomorrow morning.” And because opens and clicks arrive quickly…

Read more

If you’ve run a lot of email A/B tests, you’ll recognise the feeling: you test two subject lines, the results come back “too close to call,” and you either pick one anyway or the platform picks for you.

The industry calls this “inconclusive”. I call it normal.

Because the problem usually isn’t the subject line. The problem is that the test wa…

Read more

Email isn’t a marketing channel. It’s the operating layer almost every system assumes will be there.

Not your marketing. Your business!

Because email isn’t a tactic. It’s the one layer everything else quietly leans on: identity, receipts, resets, notifications, disputes, contracts; the stuff that keeps commerce moving even when your ads are off…

Read more

The Times ran a story this week about a business built on Instagram that effectively vanished when the account disappeared.

It’s easy to read that as “an Instagram problem”, a creator problem, or an edge-case about hacks and customer support queues.

But the useful question isn’t “could this happen on Instagram?”

The useful question is: could …

Read more

Whether you are trying to win budget, funding or new business, the first challenge is no longer just persuading the decision-maker. It is making sure your proposal survives the AI layer that decides whether the decision-maker ever sees it properly.

A proposal used to have one primary job.

It had to persuade the decision-maker.

That was never e…

Read more

This is where my writing now lives.

For years, most of it lived on LinkedIn: The Frequency File, Thinking Allowed, and the shorter posts in between. LinkedIn helped the ideas travel. It introduced me to people I would not otherwise have met. It proved there is an audience for senior-level thinking about email, attention, identity, attribution, AI, platforms and digital behaviour.

But a feed is not a home for the work.

Thinking Delivered is the home.

I started in publishing, moved online early, and have spent most of my career around the systems that shape how people see, remember, ignore, trust and buy. That is still the territory.

Email is part of it.

So is attribution.

So is identity.

So is AI.

So are platforms, media, trust, deliverability and the mechanics of attention.

The common thread is not the category.

It is the machinery underneath.

I write for operators, founders, senior marketers and people who make decisions where the usual best-practice answer is too small. The aim is to make digital behaviour more legible: what changes behaviour, what creates leverage, what distorts judgement, what survives platform shifts, and what the numbers are really telling you.

What is already here

This is not a publication starting with an empty archive.

I have written more than 200 articles over the years. I have started by bringing across a core archive of the most-read and most useful pieces, including the top articles from the last year.

That archive will keep growing.

The point is not to dump everything in one place. It is to organise the work so that the ideas compound: email as publishing, opens as signal, frequency as strategy, spam as retrieval, attribution as distortion, platforms as incentive systems, identity as infrastructure.

There are also years of podcasts, interviews, videos and clips that will become part of the library over time.

So subscribing is not only about the next post.

It is access to the body of work as it becomes properly structured.

Where to begin

Start with these pieces.

Why I’m Moving Here

Why I moved the work to Substack, and why the writing needed a home of its own.

Email Opens Matter

Why the industry’s correction on open rates went too far, and why opens remain one of the largest signals in email.

Email Is the Only Channel That Can Never Be Replaced Why email isn’t a marketing channel at all — it’s the identity layer every digital system depends on.

What Inbox Placement Looks Like in 2026 (The Spotlight Is Shifting)

The new rules of inbox placement — and why the centre of gravity is moving from authentication to behaviour.

And sometimes I don’t talk about email at all — because the same behavioural and incentive structures show up everywhere.

Gambling Ads Are Everywhere — The Case for a Frequency Cap Why gambling ads persist despite every reform — and how optimisation engines escalate pressure on vulnerable users until frequency becomes the real harm.

Together, those pieces give you the shape of the publication: email as infrastructure, metrics as signals, attention as behaviour, and marketing as a system rather than a set of tactics.

What subscribers get

Subscribers get the full publication — the complete thinking as it’s published.

That includes:

• All new essays

• The full briefings, not the public excerpts.

• 50% discount on all strategy papers

• Access to the growing archive

• All past essays, notes, and papers in one place.

This is the complete reading experience — the full intellectual output.

Founder membership

Founder Members get everything subscribers receive, plus a direct line to me.

That includes:

• A 90‑minute advisory session
  Bring the problems no one else can solve — the session is yours.

• Early access to strategy papers and thinking
  You see the work before it’s released publicly or to subscribers.

• The ability to bring specific questions, problems, or situations into the conversation

  Founder Members can ask for clarification, challenge assumptions, and request deeper exploration.

This tier is not “more content.”
It’s proximity.
It’s access.

Subscribe if you want the full archive, the new briefings, and a clearer way to think about the systems behind marketing and digital behaviour.

Subscribe if you want the work in one place, rather than scattered across feeds, newsletters, posts, interviews and old links.

Subscribe if you want thinking that treats email, attention, attribution, identity and AI as connected systems.

That is what I am building here.

Dela Quist: Thinking Delivered.

A home for the work — and for the people who want to think with it.

Subscribe now

Auto‑renewal pricing is one of the oldest tricks in the book.

Insurance companies do it. Broadband providers do it. Streaming services do it. LinkedIn does it too.

It works like this:

• New customers get the best price

• Returning customers get a decent price

• Auto‑renewing customers get the worst price

LinkedIn Premium follows this model almost perf…

Read more

Part Two: The Future of Privacy Will Be Won or Lost in the Gap Between a User’s Perception and the Reality of How They Are Targeted

In Part One I covered the €325M fine Google received for running ads inside Gmail that looked like emails but weren’t and for doing so without proper consent. For some, this was confirmation of what many ha…

Read more

Last week, Google and Gmail quietly stepped back from owning reputation. It washed its hands of IP and domain reputation via Postmaster Tools, effectively saying: “We’re not judging anymore.” That move, as I wrote last week - looked a lot like a strategic legal dodge. The reputational theatre was over. Delivery, they signalled, would now …

Read more

Many brands already treat acquisition sources differently.

A search signup doesn’t behave like a competition signup. A referral doesn’t behave like a discount popup. A marketplace customer doesn’t behave like a direct subscriber.

Thats because the context that brought them in is different, and context shapes what they expect next.

It’s why Chat…

Read more

Let’s face it: getting laid off is a commercial event. It’s not a betrayal or a punishment. You were part of a cost equation, and now you’re not. What you do next will determine whether that decision ends your momentum or unlocks it.

Forget the language of crisis. You’re not broken. You’re now running a one-person company with a single …

Read more