Why do I doubt my own memory and perception around certain people?

Can Another Person Change What You Remember Without Lying to You?

Yes. This is the single most useful fact on this page, and it is the one that explains your experience without requiring anybody to be a villain.

Gabbert, Memon and Allan ran the study in 2003. Pairs of people watched what each believed was the same video of an event. In fact each member of the pair saw a slightly different version, containing details that only one of them could possibly have seen. Then they discussed what they had watched, the way anybody would. Afterward they were tested individually.

Around seven in ten of the people who had discussed the event reported details they had never seen, having picked them up from their partner. Not because they were credulous. Not because anybody pushed them. They talked about it, and their memory absorbed material from the conversation and filed it as their own.

Nobody in that room was lying to anybody. There was no manipulation, no advantage to be gained, no relationship history, and no motive of any kind. Two people talked and one memory reshaped the other.

This effect has a name, memory conformity, and a laboratory literature of its own. The general finding is stable: when two people who witnessed the same thing discuss it, their accounts converge, and the convergence runs toward whichever account is delivered with more confidence rather than toward whichever one is more accurate.

Now put that alongside the fact that you talk to this person constantly, about things that matter to both of you, and that a great many of those conversations are precisely conversations about what happened. You are not an unlucky person with a defective memory. You are a person whose memory is in daily contact with another account, under exactly the conditions that produce this effect, with somebody who happens to state things more definitely than you do.

The caveat is the same as before and it is a real one. That work was done in laboratories, with staged videos, minutes after the event, between strangers with no stake in the answer. The direction transfers well and has been reproduced. The magnitude, over eleven years of a marriage, about a conversation in a hallway, is not something anybody has measured, because nobody can.

There is a further wrinkle worth knowing, and it is that retelling does its own work independently of anybody else. An account that gets told repeatedly gets shorter, gets a structure, and quietly loses the parts that complicate it. That mechanism is described in more detail on the page about bringing a third person into things, and it is worth knowing here because it means both accounts in a disagreement have been drifting, not only yours.

How Good Is the Evidence Behind Any of This?

Weaker in one half and considerably stronger in the other, and the split is worth knowing because almost every page on this subject presents the two halves as though they had the same standing.

On the term itself, the evidence is thin. There is almost no direct empirical literature on gaslighting as a phenomenon. What exists is clinical description, case material, sociological theory and a very large amount of opinion writing. There is no instrument, no validated measure, no agreed criteria, and no body of controlled studies. This is genuinely unusual for a word in this much circulation. When somebody tells you that research shows that gaslighters do a particular thing, the honest question is which research, and the honest answer is usually that a clinician wrote it down and other people repeated it.

On memory distortion, the evidence is strong. Reconstructive memory, misinformation effects, memory conformity and false recall have all been reproduced many times, in many laboratories, across decades. The direction of these findings is about as settled as anything in this area gets.

But that strength comes with a boundary that popular writing routinely ignores. Almost all of it was produced with staged materials: filmed events, word lists, slide sequences, sessions measured in minutes, mostly student volunteers, and no relationship between the participants. Nobody has run a study on eleven years of a marriage. Nobody could, and nobody would get it past an ethics board. So the direction transfers and the magnitude does not, and any page that tells you how much of your memory is being altered by your partner is telling you about its own confidence.

And the checklists have nothing behind them at all. The lists circulating online, promising to tell you whether you are being manipulated, have no published measurement properties, no validation study, and no known false positive rate. They are written so that ordinary friction and deliberate deception score identically, and so that almost nobody can come out negative. That is not a test. A person genuinely being deceived deserves better than a quiz, and so does a person who is simply in a difficult conversation with somebody who remembers things differently.

What survives all of that is modest and it is enough to work with. Memory is reconstructive. Confidence is a poor guide to accuracy. Another person's account can reshape yours with no intent to deceive, and the more confident account tends to win. Those four sentences explain a very large share of what happens to you around this person, and none of them requires anybody to have done anything to you.

Are Online Gaslighting Checklists Validated or Measured in Any Reliable Way?

No. Not one of the lists circulating under that heading has published measurement properties, and it is worth being specific about what is missing rather than leaving it at a general complaint.

A set of questions earns the word validated by doing three things in public. It has to be stable, so that the same person answering a week apart lands in roughly the same place. It has to be shown to be picking up the thing it claims to pick up rather than something adjacent, like general unhappiness or a bad month. And it has to be checked against an outside answer, so that somebody can say how often it says yes when the honest answer was no. Those three properties take years to acquire, and they get published alongside the questions, so that anybody can see how badly a set of questions performed as well as how well.

None of the online lists carries any of that. There is no stability figure, no comparison against anything outside itself, no error rate in either direction, and no published sample the items were developed on. What you are answering is a set of statements somebody wrote down because they sounded right.

The way they are written makes it worse. Read the items closely and most of them describe things that happen in ordinary difficult relationships. You feel confused after conversations. You apologize a lot. You check your version of events before you speak. Somebody who is genuinely being deceived will say yes to all three. So will somebody in a rough patch with a person who remembers the week differently and states it firmly. A list on which those two people score the same is not separating them, and a list almost nobody can come out of negative is not sorting anything at all.

There is a harder problem underneath the missing numbers, and it is the reason a properly built version would be difficult even for somebody willing to do the work. To check any list against reality you would have to know, independently, which of the people filling it in were actually being deceived. Nobody has that. Intent is not visible from outside, the other person is unlikely to confirm it, and the whole difficulty being described is that the facts are in dispute. With no outside answer available, there is nothing to hold the questions to.

What that leaves is smaller than a verdict and considerably more use than one. The safety questions at the top of this page work differently, because they are four specific facts about your circumstances and they either apply to you or they do not. Below that line, the thing worth doing is the one at the foot of this page, which produces evidence about your own accounts rather than a score about somebody else's motives. A list that returns a label is answering a question nobody can currently answer. Four lines written before a conversation are answering one you can.

Why Do Two People Doubt Their Own Memory for Completely Different Reasons?

Because the experience is the surface and the cause is somewhere underneath it, and two people can produce an identical description of that surface from entirely unrelated causes.

Put two people side by side. Both hesitate before stating what happened. Both check messages before speaking. Both say "I might be remembering this wrong" more often than they say anything else. From outside they are indistinguishable, and any advice written for one of them will be worse than useless for the other. This is the whole reason generic writing on this subject has such a poor record.

Here are the five roots, in the order they should be checked. The order is not arbitrary, and the first one is not a root to work on here.

Check the first row honestly and stop there if it applies to you. Work downward only if it does not. Roots also combine, and the commonest combination by a distance is the second and the fifth: ordinary memory drift landing on somebody who was trained early never to hold their own version of events.

What If Two Honest People Simply Remember Differently?

This is the commonest root by a wide margin and the one that gets misread most often, because the reader has been treating a normal property of memory as evidence of a personal defect.

Here is the shape of it. You and somebody else were in the same room. Your attention was on different things, which is ordinary. The parts that mattered to each of you were different, which is also ordinary. And in the weeks since, each of you has told the story to other people who reacted to particular parts of it, which quietly reshaped both versions in different directions. Two sincere, incompatible accounts, produced by nothing more sinister than two functioning memories doing what memories do.

That first part of it has been measured directly. A 1978 study in the Journal of Verbal Learning and Verbal Behavior, published as "Recall of previously unrecallable information following a shift in perspective", gave people the same written description of a house along with a point of view to hold while reading it, one group as somebody thinking of buying the place and another as somebody planning to rob it, and what each reader could produce afterward followed the assigned point of view rather than the page, with details missing from one version surfacing later when the reader was asked to switch. The limit belongs alongside it, because that was a written passage read by students in a room, minutes rather than months apart, with nobody arguing and nothing at stake, so it shows that a point of view decides what you can retrieve without saying how much of a disputed Thursday it accounts for.

The tells for this root are unusually clean.

It happens with people who have nothing to gain. A friend, a sibling, a colleague with no stake whatsoever in whose version wins. If your record comes apart with those people too, at least sometimes, you are looking at ordinary drift rather than at anything being done to you.

The disagreements are about details rather than whole events. You both agree the conversation happened, in the hall, that week. You disagree about the day, or the number, or the exact wording. Systematic rewriting tends to remove whole events. Drift moves the furniture inside them.

And nobody comes out of the disagreement ahead. If the version that wins produces no advantage to anybody, the machinery grinding out these disputes is probably just memory.

What changes when this is your root is more than it sounds. The question stops being who is lying and becomes what actually happened, which is a question two people can be curious about together. It also stops being an emergency. Being wrong about a Thursday becomes a thing that happens to people rather than a piece of evidence about your reliability as a person.

One more piece belongs here. The version of this that plays out between siblings about a shared childhood, where two people remember two different parents, has its own mechanism and its own page. That one involves an account supplied by somebody else early, which is a different problem from the ordinary drift described here.

What If You Were the One Who Was Wrong, Once?

Then a single real error is still teaching, years after the lesson stopped being useful.

This root has a distinctive signature: you can name the occasion. Often to the month. You were sure about something, you said so clearly, possibly in front of people, and you turned out to be flatly wrong. Perhaps it cost something. Perhaps somebody was hurt by it, or you looked foolish in a way that stayed with you, or a decision got made on your version and the decision was bad.

What a mind takes from an event like that is almost never proportionate. The reasonable lesson would be narrow: on that class of thing, under those conditions, check first. The lesson that actually gets installed is broad: my certainty is not to be trusted. And a broad rule costs nothing to keep running, so it goes on running, applying itself to situations that have nothing in common with the original.

The rule survives because it is never tested. Every time you defer instead of stating your version, you avoid the possibility of being wrong again, and you also avoid the evidence that would have shown you the rule was too big. Being right and staying quiet about it produces no correction, because nobody ever finds out. So the incident from four years ago remains the most recent data point, indefinitely.

There is a particular version of this involving a person rather than an event. You trusted somebody, your read of them turned out to be badly wrong, and the conclusion you drew was about your perception rather than about them. That version is handled properly on the page about not trusting your own judgment, which owns the wider faculty. What belongs here is the narrower case, where what you stopped trusting was specifically your record of what happened rather than your read of people.

The useful move for this root is to say the thing out loud. Name the incident, to yourself, in one sentence, with the date. Then notice how much of your current behavior is being run by it. People are frequently startled to find that a rule governing several conversations a week is resting on one afternoon in a year they have to stop and calculate.

Where Does This Hide When You Would Not Call It Doubting Yourself?

In half a dozen places that most people would file under being reasonable.

There is the pre-emptive hedge. "I might have this wrong, but." Attached to a sentence you have no doubt about at all, inserted purely to make the retreat cheaper if it is contradicted.

There is the silent verification. Scrolling back, checking a calendar, reading an old message, all done before you speak so that nobody can see you doing it. The checking itself has become invisible, which is why the doubt underneath it never gets named.

There is the abandoned correction. Somebody states something you know to be inaccurate, and you let it stand, and the reason you give yourself is that it does not matter. Sometimes true. Sometimes it is a much older reflex arriving in the shape of a reasonable decision.

There is the pre-writing. Composing what happened, in a note, before a conversation, so that you can hold it under pressure. Very few people describe this to anybody, because it sounds strange said out loud, and a surprising number of people do it.

There is the borrowed account. Somebody else says how it went, and you find yourself adopting their version quite easily, and later cannot reconstruct what you thought before they spoke. This one is memory conformity operating in plain sight.

And there is the general softening. Over time, you make fewer definite statements about anything. Not only around them. It leaks, because a habit built for one relationship does not check whose company it is in before running.

Why Does Every Sensible Fix for This Fail?

Because each of them addresses a conclusion, and the thing they are aimed at is not a conclusion.

Deciding to trust yourself more fails immediately, and the reason is mechanical. The doubt arrives in the seconds after a confident contradiction, before any reasoning has taken place, in roughly the way a flinch arrives before you have decided anything about the object coming toward you. By the time you notice it, it has finished. A resolution made on Sunday evening is being made by a part of you that does not get consulted on Tuesday morning.

Checking harder fails for a different reason, which is that it works. You look up the message, you find the date, you are right. And nothing changes, because the doubt was never a question about that Tuesday. Winning individual events one at a time leaves a standing condition of the faculty exactly where it was, which is why people describe being conclusively vindicated and feeling the same collapse at the next disagreement three days later.

Arguing about it fails in a way that is worth understanding rather than just noting. A disagreement about what happened has no available referee. Both accounts are sincere, neither can be produced, and the conversation converts into a conversation about who is more reliable, which the more confident speaker wins by default. You will lose that one every time regardless of who was right about the appointment.

Documenting fails for reasons of a different kind. It moves your confidence out of you and into a file, so that on the day the file has a gap in it, you have less than you started with. It converts a relationship into a case you are building. And it changes what you are doing in conversations, because a person who is collecting has stopped simply talking. This page does not recommend it, and the exception is the safety situation named at the top, where that decision belongs with somebody who knows your circumstances rather than with a website.

And being told that you are being gaslit fails most interestingly of all, because it feels enormously like help. It supplies a villain, it explains everything, and it costs nothing to accept. If the word fits your situation, none of the relief it brings actually changes your circumstances. If it does not fit, you have swapped a difficult uncertainty for a comfortable story, and the story will hold you in place for years.

What Can I Do Without Confronting Them or Secretly Recording Conversations?

Nothing that helps here needs their cooperation, their agreement, or a recording device, and the one thing that does help is smaller than any of those. It is easiest to reach through a different question: which part of this did you actually choose?

Almost none of it, and the exception is small enough that people dismiss it.

The doubt arrives before deliberation is available, which puts it outside anything you could have chosen, and there is no point in arguing with it. The checking is barely chosen either, because it starts before you have framed a decision to check. Conceding to end a disagreement is a choice in the technical sense and it is made under conditions that make the outcome close to inevitable.

There is one thing in the whole sequence that you choose freely and that costs nothing, and it works for every root above.

Before a conversation you expect to be contested, write four or five lines about what you remember. In the present tense. With today's date. Then put it away.

That is the entire exercise. Three rules make it work, and each of them matters.

It is for you and it is never produced. You are not building a case, you are not going to read it out, and you are not going to reference it in a conversation. The moment it becomes something you might show somebody, it stops being what this is and becomes documentation, which is a different activity with different consequences and is not what is being suggested here.

It goes in before the conversation, not after. The drift happens during the exchange. A version written afterward has already been through the thing that alters it. Writing it first is not about being stronger, it is about your account existing somewhere the conversation cannot reach.

You do not act on it at the time. No producing it as evidence, no saying that you wrote it down, no using it to win anything. It sits there. You read it back later, alone, when the temperature has dropped.

Why this serves every root is the part worth understanding. If ordinary drift is your problem, a month of these will show you drift, in your own accounts, which is oddly reassuring rather than alarming. If it is a confidence gap, you will see your version surviving intact right up to the moment it meets theirs, which locates the effect precisely. If a single old error installed a rule, you will find that your accounts are mostly accurate, and the rule will start looking as oversized as it is. And if you were never allowed your own version growing up, this gives the faculty the one thing it never had, which is a record it made itself, without permission from anybody.

There is one more reason it belongs here rather than any of the more energetic suggestions. It requires nothing at all from the other person. No conversation, no confrontation, no cooperation, no announcement. Everything else people are told to do in this situation depends on somebody else behaving in a particular way, which is exactly what you cannot count on.

What Does the Settled Version Look Like on an Ordinary Week?

Considerably less dramatic than people expect, and that is the useful part.

The disagreements still happen. You still sometimes turn out to be the one who had it wrong, because that is a thing that happens to people who have memories. What changes is that a contradiction stops taking your entire record down with it. Somebody says Tuesday, and you notice that you had Thursday, and both of those facts get to sit in the room at once without one of them having to be destroyed.

The hedging thins out. Not because you have become assertive, but because you have stopped needing the escape route in front of every sentence. You say what you remember and you say it plainly, and if it turns out to have been wrong you are wrong the way anybody is wrong.

The checking gets less compulsive. You still look things up, because looking things up is sensible. What goes is the checking done in secret before you are allowed to speak.

You become able to say "I remember that differently" and then stop talking. That sentence, on its own, without a follow-up and without a defense, is most of what the settled version consists of. It leaves the matter unresolved, which turns out to be a state a person can live in comfortably once they stop treating every unresolved event as a verdict about themselves.

And the general softening reverses. Other people notice this before you do. Somebody will tell you that you seem more definite lately, about matters that have nothing to do with any of this, and you will not immediately know what they mean.

What Is Worth Doing This Week, Before You Say Anything to Anybody?

Four lines and a date, and nothing else.

Before the next conversation you expect to be contested, open a note. Write what you remember, in the present tense, with today's date on it. Do not show it to anybody. Do not mention that you wrote it. Do not use it in the conversation. Put it away and go and have the conversation exactly as you would have anyway.

Do that for a month. Then read them back in one sitting.

You are looking for one thing: whether your accounts drift, hold, or were never the problem. That answer is not currently available to you, and it has not been available to you for years, because the design of the difficulty means the evidence gets destroyed by the very conversation that would have produced it.

If what landed hardest here was the research, the writing will probably be enough on its own, and you can stop there. If what landed hardest was something else, the recognition that this predates the person you were thinking about, or that you can date the afternoon that installed the rule, or that you have not stated anything definitely to anybody in about six years, then the thing worth working on sits underneath the doubt rather than inside it. Our Sacred Valley seminar works at that layer. There is a temptation in this whole subject that is worth naming, because it is the reason so many people read about it for years and change nothing: a lot of this work is basically looking for explanation. In Florian's words: "We don't want to be people who are just making ourselves feel good by having a good reason." A word that explains everything is a comfortable place to stop. We start somewhere less comfortable, which is what the behavior actually does for you, and then what is standing in the way, and then a plan you are held accountable to afterward, back in the same hallway with the same person. Everybody is screened before arriving, which keeps the group small enough that somebody who has spent four years hedging every sentence gets worked with properly rather than handed a worksheet. One limitation worth stating plainly: exact dates, the price, and the final cap on group size are still being settled, so anybody comparing programs on those numbers this week cannot yet compare ours. Ask every program you are weighing the same questions, this one included, and decide from what comes back.

The four lines cost you nothing and are worth having either way.