Why do I keep doing the thing I said I would stop, and then explain it to myself?

Does Self-Perception Theory Disprove Cognitive Dissonance?

No, and it has never been ruled out either, which is an unsatisfying answer that happens to be the accurate one.

The rival account was set out in 1967, in Psychological Review, in a paper called Self-perception: an alternative interpretation of cognitive dissonance phenomena. Its claim is that nothing uncomfortable has to be happening inside you at all. People have poor access to their own attitudes, so they work them out from the outside, by observing what they did and the conditions it happened under, then inferring what they must have thought in order to have done it. I am naming the paper rather than its author here, because the surname belongs to two researchers whose work has nothing to do with each other, and the other one is cited elsewhere on this site for something entirely unrelated.

The demonstration that made it hard to wave away was a simple one. People who merely read a description of the 1959 study, and who therefore had no discomfort of their own to resolve, predicted the participants' reported attitudes from the outside and got them right. If an observer with nothing at stake can reproduce the result, the inner state may not be the thing producing it.

What eventually happened was a division of territory rather than a verdict. A 1977 experiment in the Journal of Experimental Social Psychology found each account fitting a different range: where the behavior sits well outside the position you already hold, the discomfort account predicts better, and where it sits close to it, the inference account does. Nothing since has improved on that decisively, and the argument went quiet mostly because the field found other things to do.

On the evening this actually happens to you, the two accounts agree about everything you can observe. The belief ends up where the behavior already is. What they disagree about is the story of how it got there, and if anything the inference account is the more uncomfortable of the two, because it describes no repair being run on your behalf at all, only a person reading their own conduct and drawing the obvious conclusion from it.

Why Does It Sound Thinner When You Say It Out Loud?

Because inside your head it has never once had to survive an audience that was not already on its side.

This has been measured in settings nothing like yours, and the shape of the result is the useful part. Work on the illusion of explanatory depth, published in Cognitive Science in 2002, found that people rate their understanding of ordinary objects highly and then rate it far lower once they have been made to spell out how the thing actually works, and a 2013 paper in Psychological Science, called Political extremism is supported by an illusion of understanding, reported the same deflation when people had to explain how a policy would operate, while simply listing their reasons for holding it left them where they started. I am naming the papers rather than their authors, since another page on this site already carries that attribution. Neither study went anywhere near a personal excuse and the 2013 shift was modest in size, so take the pair as a description of the mechanism rather than proof about your own evening: reasons hold up when spoken, an account of how something came about generally does not, and a friend who asks how rather than why has asked for the second kind without knowing it.

An explanation held privately gets tested against nothing. It arrives, it fits, the discomfort stops, and that is the entire quality-control process. There is no version of you standing slightly back and asking whether the third clause is doing work the first two failed to do. The listener and the speaker are the same person and that person wanted the matter closed.

Said out loud, three things happen at once that never happen internally.

It has to be sequential. Inside, the reason exists all at once, as a kind of atmosphere. Out loud it has to be laid end to end in a specific order, and the order exposes which parts are load-carrying and which are decoration. People frequently hear themselves reach a clause and know, mid-sentence, that they are about to skip it.

It arrives in front of somebody who has the earlier version. If you told this person about the rule when you set it, they are holding both, and you can feel them holding both even if they say nothing at all. The gap becomes audible to you before they respond.

And it takes on the specific quality of being recited, which is the tell people report most often. That quality is information. It means the account was already finished before the conversation started, and a genuine reason produced in the moment does not sound like that.

None of this makes the explanation false. Nothing about it changed on the way out of your mouth. What changed is that its only support, an audience that agreed with every step, was removed, and it turns out that support was carrying more than you thought.

Is It a Lie?

No, and getting this wrong will cost you the whole page.

The material is true. It genuinely was a long week. You genuinely were not driving. One of them genuinely does not matter very much on its own. If any of those were false the explanation would not work, because you would catch it, and the whole point of the repair is that it passes inspection by the only inspector present.

What is inaccurate is smaller and much harder to see. The explanation makes a claim about causation that it never states out loud. It presents itself as the reason the thing happened. It is actually a set of true facts that were available in the environment, selected afterward because they fit, and assembled into a shape that resolves a contradiction. The facts are real. The causal claim is retrofitted, and the retrofitting is invisible because nothing false was used to do it.

One study separates sincerity from accuracy about as cleanly as this can be done. Johansson and colleagues reported in Science in 2005 that people who chose the more attractive of two pictured faces, and were then handed back the face they had rejected by sleight of hand, mostly failed to notice the swap and went straight on to give confident, specific reasons for a preference they had never expressed. Those reasons were not dishonest, only produced in good faith and at speed about something that had not happened. Roughly a quarter of the switched trials were caught, and it was a card task with strangers' faces rather than a life, so what it shows is that a sincere reason can be built for the wrong event, never that yours always is.

This distinction is the reason the standard advice does nothing. Being told to stop lying to yourself lands on somebody who is not lying, and the instruction bounces, correctly. You examine the explanation, find it accurate, and conclude that this particular framework does not apply to you. It applies exactly as much as it did before, because accuracy was never the issue.

There is a related trap worth naming, since the word tends to invite it. The repair is not a defect in you. Anybody without it would be intolerable to live inside, spending their days accumulating unresolved contradictions with no way to close any of them. What you are looking at is a normal function running normally. The question worth asking is what it has been quietly buying, not whether you should have it.

Why Can You Never Catch the Standard Moving?

Because the movement is sized precisely so that it cannot be caught, and the sizing looks deliberate.

It is never a reversal. Nobody goes from I do not do this to I do this. The adjustment is closer to a notch, roughly the amount required to accommodate one instance and no more. Weeknights becomes weeknights that were normal. Nothing after eleven becomes nothing after eleven unless something is actually happening. Each edit is small enough to look like a clarification of what the rule always meant, and clarifications do not trigger any alarm, because refining a rule is what a thoughtful person does.

There is a second reason, and it is structural. The rule lives in the same place that does the repairing. When the repair runs, it edits the rule in the same pass, so at no point do you hold both versions simultaneously. You cannot compare a thing to itself. By the time you go looking, only the current version exists, and it feels like the version you always had.

The only way anyone ever sees it is by comparison against something outside their own head. A note written years ago. Somebody quoting you back to yourself. A photograph of a page. That is why people discover this in a specific and consistent way: they find something they wrote, read it, and have a sharp physical reaction to how far it is from where they now are, with no memory of any point at which they agreed to move.

Nothing about that is a failure of attention. There was no moment to pay attention to. The whole thing happened between the pours.

What If This Is Just the Ordinary Repair Everybody Runs?

Then very little is wrong, and this deserves naming first and generously, because it is the most likely answer and the one nobody writes.

Every person alive does this, continuously, about a great many things. The capacity to reconcile what you did with what you hold is standard equipment and it is doing far more good than harm. It lets people act without being paralyzed by the gap between their conduct and their values, and that gap is permanent, universal, and not going anywhere. Somebody who felt the full contradiction every time would get very little done and would not enjoy any of it.

If you have arrived here because you noticed yourself producing an explanation and it disturbed you, the honest reading is that your attention is unusually good and nothing much is wrong. The ordinary condition is running this repair several times a day for decades and never once observing it. Noticing puts you ahead of most people.

The tell for this root. The behavior itself is genuinely minor, the cost is genuinely absorbed by you alone, and if you subtract the explaining, what is left does not amount to a problem. The distress is coming from having seen the machinery rather than from anything the machinery has done.

What this root needs. Substantially nothing. Possibly a written rule, out of curiosity rather than concern, on the one behavior you are least sure about. If it holds, you have your answer and can stop reading about this.

The reason to name this root first is that the other four are only worth investigating once this one has been ruled out, and a page that opens with the expensive versions will convince people they have a problem they do not have.

What If What You Are Defending Is a Sentence About Yourself?

Then the behavior was never the point, and this explains the thing that most puzzles people: why the explanation is so disproportionate to what happened.

Some rules are attached to a description you rely on. I am someone who does not do this. I am not the sort of person who lets that slide. I am reliable about exactly this one thing. The description does real work. It may be one of a small number of things you are confident about, and in some cases it is holding up an area where the evidence is otherwise mixed.

When the behavior contradicts the description, what takes the hit is the description, and the description is much more expensive than the evening. So the repair works far harder than the event warrants. This is why people find themselves producing an elaborate account of something that objectively barely happened, and why the intensity feels strange even at the time.

The tell. Disproportion, and a particular kind of it. The explanation is longer, more urgent, and more thorough than the act could possibly justify. There is often a second stage where you rehearse it, hours later, against an objection nobody made. And the effort scales with how central the description is rather than with how bad the behavior was, which is why small acts sometimes generate enormous repairs and larger ones sometimes generate none.

A second tell is the shape of the explanation. It tends to be about who you are rather than about what happened. Not I did this because the week was bad, but this is not the kind of thing I do, therefore this instance must be a special case, therefore here is the special case. The special case is generated to protect the sentence.

What this root needs. The uncomfortable question is what the description is holding up. A sentence about yourself that has to be defended this hard is usually load-carrying somewhere else, and finding out where is a different job from managing the behavior. The related pattern where a single event gets converted into a verdict about your whole character runs on the same machinery pointed the opposite way, and is worth reading if the sentence you are defending is one you half suspect is already false.

Why Do I Explain Away Missed Commitments to My Boss Even When I Meant to Keep Them?

Because you meant it and you missed it, and only one of those two facts is still soft enough to edit.

The intention was real, and that is worth settling before anything else, because most people arrive at this convinced they must have been insincere at the moment they agreed. A 2006 meta-analysis in Psychological Bulletin pooled forty-seven experiments that had successfully shifted people's intentions and found that a medium-to-large change in intention produced only a small-to-medium change in what those people actually did. Meaning it and doing it are that loosely coupled in ordinary people under ordinary conditions. Most of those studies were about health behavior over short windows and none of them were about work, so take the number as a floor under the general point rather than a measurement of your own week. I am naming the analysis rather than its authors, since another page on this site already carries that line of work.

Then the miss happens and the repair runs exactly as described above, with one difference that changes how the whole thing feels. At work the account has to be said out loud, to somebody who was waiting for the thing. Said to a friend with no stake in it, an explanation deflates. Said to the person it cost, it gets rebuilt instead, because it now has to be acceptable to them, and the acceptable accounts are the ones pointing at causes outside your control: the other project, the illness, the person who never sent the file. Another page carries the research on how people select their excuses, and the short version is that the material is usually true and the selecting is what does the work.

What quietly moves is the status of the commitment rather than your view of yourself. By the time you have explained it twice, the thing you agreed to has become a target rather than a promise, something everybody understood to be provisional, the sort of date nobody was treating as exact. No announcement was made, and you could not name the afternoon it happened. It is the same notch this page has been describing all along, applied to an agreement instead of to a rule.

The tell at work is the second person in the room. Your account will contain a claim about what they expected, and that claim will be the least evidenced part of it. They knew it was going to be tight. Nobody was actually waiting on it. That is the fourth root above, arriving in a place people do not look for it, because the cost has landed on somebody who is paid to absorb it.

What to do is the same thing the rest of this page recommends, moved earlier in the sequence. Write the commitment down in the words you used and with the date you gave, at the moment you give it, somewhere nobody else has to see. Six weeks later there is a version from before, and whether you keep what you agree to becomes a fact instead of an atmosphere.

If your version of this is mostly about agreeing to more than any week could hold, the page on promising too much fits it better than this one does.

Why Does Deciding to Be More Honest With Yourself Fail?

Because it aims at the wrong moment and asks for a thing that is not missing.

Honesty is not the deficit. The explanation is made of true material, which is the entire reason it works, so an instruction to be more honest arrives at a door that is already open and finds nothing to do. People run this instruction for a while, examine their explanations, find them accurate, and reasonably conclude the framework does not describe them.

The second problem is that a resolution to be more honest is made of exactly the same substance as every other resolution, which means the repair can process it like anything else. The first time honesty is inconvenient, an account will arrive explaining why this particular case does not require it, and the account will be good, because they always are.

The third problem is timing. Every instruction of this kind points at the moment of the behavior, and at that moment there is nothing available to work with. The pressure has not arrived yet, so there is nothing to notice, and by the time there is, the repair has already completed. Aiming at the one part of the sequence you have no access to is why so much effort here produces so little.

Notice what is being ruled out. Watching for your own rationalizations, which fails for the reasons above. Announcing intentions to somebody so they will check on you, which turns a private rule into a performance and produces explanations aimed at them instead of at you. And any general resolution to be more truthful, which is a rule with no number in it, and rules with no numbers are the most movable objects on this page.

What Should I Write Down So I Can Catch My Standard Changing?

Catching a standard while it moves needs a version of the rule from before it moved, and creating that version is the only part of this whole sequence you actually choose.

Writing the rule down, in specific words, before the next occasion. That is the whole list.

Everything else described here happens to you. The pressure arrives on its own. The explanation arrives on its own, complete and unrequested. The standard adjusts without consultation. None of those are available to be decided about, and treating them as though they are is what makes people feel they are failing at something.

One thing is chosen, and it is chosen on an ordinary afternoon when nothing at all is happening, which is exactly why it never feels urgent enough to do.

Why written and not remembered. A remembered rule is stored in the same place that performs the repair, so it gets edited in the same pass, and you never hold both versions at once. A written rule has a version from before. It sits outside the system that would otherwise revise it, and it cannot be quietly reinterpreted by somebody who wants it to say something else this evening, because the original wording is right there. The comparison that is permanently unavailable inside your head takes about seven seconds on paper.

What it has to contain. A number, a time, a place, or a condition. Nothing evaluative.

Two, and the third is where I stop. Not: I will drink less.

Nothing after eleven. Not: I will be sensible about my phone.

No orders on days I have not eaten properly. Not: I will stop shopping when I am upset.

Evaluative words are the mechanism by which a standard moves without appearing to. Less, sensible, reasonable, moderate, and properly all mean whatever the evening requires them to mean, and they can be reinterpreted with no sensation of anything having changed. A number cannot. That is the entire reason for the number, and it is why the rule should feel slightly too rigid when you write it.

What to do with it. Put it where you will encounter it without looking for it. Then leave it alone. There is no tracking system here and nothing at all to record. Its whole function is to still say what it said.

What happens when you break it. You get a fact instead of a fog, and that is the point rather than a failure of the method. This rule, these conditions, this many times, with the original wording still visible. That is material an actual answer can be built from, and it is the first time you will have had any. A rule that was quietly rewritten leaves nothing behind at all.

Expect the first month to mostly produce information rather than change. The information is the deliverable.

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

Quieter than people expect, and the explanations have not stopped.

The reason still arrives. That does not change, and anybody promising otherwise is selling something. What changes is what happens in the few seconds after it arrives. There is a small recognition, without drama attached, that this is the familiar one. Not a confrontation, not a moment of honesty, just the ordinary recognition of something you have met before.

The rule is on paper somewhere and it says what it said in March. Sometimes you have kept it and sometimes you have not, and both are ordinary. The difference is that you know which, and you know roughly how often, because the wording has not moved underneath you.

On one or two things, the rule got retired deliberately, after the one-minute test produced a shrug and somebody else's voice. Nothing has gone wrong as a result. The small weekly pulse of self-reproach attached to that rule is gone, and what it was costing only became clear once it stopped.

Where somebody else was paying, you asked. The answer was less bad than the guess in one case and considerably worse in another, and both were useful, and neither could have been arrived at from inside your own account.

And the standard sits still. That is the thing you notice a year out, reading something you wrote. It says what you remember it saying. There is no jolt.

What Is Worth Doing This Week, Before You Decide Anything?

Two things, both private, neither of which commits you to changing any behavior at all.

The one-minute test, on one rule. Pick the rule you break most often. Write it in a sentence. Then write why it exists in a sentence, with no should in it and no other person named. Read the second sentence back and listen for whose voice it is in. If it is yours and it connects to something you actually want, the rule is real and worth keeping. If it produces a shrug or a borrowed phrase, you have learned that you have been failing at somebody else's rule for a period of years, which is a different problem with a much shorter solution.

One written rule, with a number in it. On whichever behavior you would most like to know the truth about. Specific enough that no reading of it can stretch. Put it where you will meet it by accident. Then do nothing else for a month, and specifically do not try to notice your explanations, which cannot be done and will only convince you that you are bad at this.

Both together take under an hour and neither requires you to tell anybody anything.

Our Sacred Valley seminar works at the layer underneath both of those. What a page cannot do is tell you which of the five roots is yours, and I would rather say that plainly than imply otherwise, because the five need different things and two of them need close to opposite things. A week is enough time to find out which rule was never yours and should be retired, which sentence about yourself has been doing more structural work than you realized, and where a cost has quietly been landing on somebody who has not said so. Then we build something specific enough to survive the Thursday it has to survive, and you are held to it afterward, back in your own week and your own city. What we are after is finding a way to truly change the behavior and have it last, have it stay. Everybody is screened before arriving, which keeps the group small enough that somebody who has never once said any of this out loud gets worked with properly. One limit worth stating: 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 two exercises cost you an hour and no conversation with anybody. By the end of the month you will know something about your own standard that no amount of thinking about last Thursday was ever going to produce.