Why do I make promises I mean and then not keep them?

Were You Lying When You Said It?

No, and getting the distinction right has a practical consequence rather than a moral one.

A lie requires that at the moment of speaking you knew the statement to be false. If you meant it, that condition is not met, and this holds even though the outcome looked identical from the outside. What you produced was an accurate report of your preferences delivered in the grammar of a prediction. Those two things use the same words and do completely different jobs, and English gives you no way to mark which one you are doing.

Most people who read a page like this have already privately settled on the harsher reading, usually years ago, because it is the only explanation that has ever been offered to them. The trouble with settling there is that it produces its own behavior. If you have concluded you are unreliable, you start avoiding the conversations where promises get made, or you over-commit to prove the conclusion wrong, and both of those generate more missed promises rather than fewer.

The distinction also matters to the other person, considerably more than most people realize. Schweitzer, Hershey and Bradlow reported in 2006 that trust damaged by an unreliable act recovered when the person apologized and then behaved consistently, while trust damaged by deception did not fully recover even with the same apology and the same consistency. Deception and unreliability are repaired at different rates, and only one of them is repairable at all in the ordinary way. If the person waiting on you has quietly filed you under the wrong one, that is worth knowing, because it is the kind of thing that gets corrected by evidence rather than by protest.

Why Is the Person Who Promised Unreachable by Thursday?

Because your own decision-making treats a future version of you with something closer to the distance it uses for other people than the closeness it uses for now.

Pronin, Olivola and Kennedy reported a set of studies in 2008 in which participants were asked to allocate an unpleasant-tasting liquid for a taste experiment. People assigned their future selves substantially larger quantities than they assigned their present selves, at levels much closer to what they assigned to strangers. The same shape showed up with time: people volunteered more of their future hours than of their current ones. When you agree to something for November, part of what is happening is that you are spending a resource belonging to someone you are only loosely identified with.

Bartels and Urminsky pushed at the same joint from the other side in 2011. When people were led to see their own identity as likely to change substantially, they became noticeably more impatient in choices between smaller-sooner and larger-later rewards. When continuity was made salient instead, patience increased. How connected you feel to the person who has to do the thing turns out to be a live variable rather than a fixed fact about you.

Hershfield and colleagues added a neural version of this: thinking about one's distant future self produced activity patterns resembling those produced by thinking about another person, and the degree of that resemblance tracked real-world saving behavior. That description is reasonably well supported. The applied versions built on top of it, showing people age-progressed images of themselves to change their financial behavior, have a thinner and more mixed record in larger tests, and it is worth saying that rather than presenting the whole line of work as settled.

What all of this gives you is a correction rather than a technique. The sentence you have probably been saying to yourself, that a person who really meant it would have done it, does not survive the evidence. The evidence suggests something less flattering to everybody and much less specific to you. Meaning it happens here. Doing it happens over there, to somebody your own brain treats as partly a stranger.

Why Do Two People Break the Identical Promise for Completely Different Reasons?

Because the promise is the visible end of something, and the something varies enormously between people who look identical from outside.

This matters practically rather than philosophically. Devise a fix based on the surface and you get a fix that works for whichever person happens to share your surface, which is not necessarily you. A rule about naming times will transform the week of somebody whose problem was genuinely a forecast, and will do nothing whatsoever for somebody whose yes was ending an unbearable moment, because that person will simply name a time in exactly the same unbearable moment for exactly the same reason.

Five roots follow. Most people find one lands considerably harder than the others, and a fair number find two running together in different parts of their life. Read them for the physical jolt rather than for the argument, and pay particular attention to the second one, which is the most common and the least shameful and the one people are least willing to accept about themselves.

What If the Yes Was the Fastest Way Out of the Moment?

Sometimes the promise works as an exit from a conversation that had become difficult to stay in, with no plan inside it at all.

The shape is recognizable once described. Somebody is upset, or waiting, or looking at you with an expectation you can feel physically. The silence has gone on slightly too long. And the yes arrives, fast, before any part of you has consulted a calendar, and the moment immediately gets easier. That easing is the payoff, it arrives within about a second, and it is completely reliable.

Anything that ends a bad moment gets repeated. It does not need to work out later in order to get repeated. It only needs to work now, which it does, every time, without fail. That is the least controversial finding in the whole of behavior science and it is doing more work in this pattern than anything sophisticated.

Two tells separate this root from the others. The first is speed: the promise leaves your mouth ahead of any thought about whether it is possible. The second is what you feel immediately afterward, which is relief rather than warmth. A promise that was genuinely a plan tends to leave a small pleasant anticipation behind it. A promise that was an exit tends to leave a faint flatness, sometimes with a hint of dread underneath, and it usually gets changed subject within about ten seconds.

The promises produced this way also tend to be vague, because specificity would have required the very pause the yes existed to avoid. Soon. This week. Definitely. If your unkept promises are disproportionately the vague ones, that is worth noting.

The moment itself, and what the word is buying inside it, is covered properly on saying yes when you mean no. What belongs here is the consequence: a promise generated as an exit was never a forecast, so no improvement in your forecasting will touch it.

What If You Just Misread Who You Would Be by Then?

This is the ordinary root, it is the most common one by a wide margin, and it deserves considerably more respect than it usually gets.

Here nothing is being bought and nothing is being avoided. You wanted to make the call. You looked at your week and formed an estimate. The estimate was made by a person who felt like calling, and by the time Thursday arrived the person who felt like calling was not available, and the person who was available had a different set of priorities that were also completely legitimate. That is a prediction error about internal states, of the kind everybody makes, in a domain where nobody has ever been trained to expect it.

The evidence supports treating this generously. State prediction is poor across the board and in both directions. Nobody looks at a hungry version of themselves from a full one and gets it right. What makes promises particularly exposed to it is that the two states are separated by days rather than minutes, and that the promise is spoken in the hotter of the two.

Two tells identify this root. The first is distribution: your misses are spread fairly evenly across people you like, people you owe, and people you barely know. There is no pattern by relationship, because the mechanism is not relational. The second is specificity: your promises tend to be detailed, with dates and plans attached, because you were genuinely planning rather than escaping. People whose yes is an exit do not produce detail.

If this is your root, the most expensive thing in the whole pattern is probably not the missed calls. It is the moral conclusion you drew from them. A forecasting error repeated for twenty years, interpreted as evidence about your character, produces a quiet and permanent low-grade self-assessment that costs far more than the handful of things you actually did not do.

What If Every Yes Was Reasonable and Only the Sum Was Impossible?

This is the root that hides best, because it cannot be found by examining any individual promise.

Herrnstein and Prelec described the mechanism in 1991 under the name melioration: choices get allocated toward whichever option currently returns the most, one decision at a time, and the aggregate that results can be worse than an allocation nobody ever explicitly chose. Nothing about any single step is irrational. The whole is bad and no part of it is.

Applied to a calendar, it looks like this. Somebody asks for a coffee. Reasonable, and you say yes. Somebody asks whether you can look at their document. Half an hour, reasonable, yes. A weekend in June, a call with an old colleague, a favor for a neighbor. Every one of these, taken alone, is a thing a decent person with a normal week can absorb. There is no point at which the wrong decision is made. And then the week arrives with eleven of them in it and four of them do not happen.

The tell is diagnostic and quick. You can defend every commitment on your list individually, in detail, and you cannot explain the week. If a friend went through your calendar item by item asking which one you should have refused, you would have a good answer for each and no answer for the total.

Which means reviewing your promises one at a time will never find anything, and that is exactly what people with this root do, repeatedly, concluding each time that they simply need to be a bit more disciplined. The only place the problem is visible is the sum, written out on one page, in one sitting, and most people have never once done that.

The taking-on side of this, the hand that goes up before the week has been checked, is volunteering for everything. What belongs here is the consequence for promises specifically: when the sum is impossible, something has to give, and what gives is always whatever had no time attached to it, which is almost always the promise made to somebody who loves you and will not chase.

Which Root Is Yours, Judging Only From This Month?

Look at what you already have rather than at how you feel about yourself, which is the least reliable instrument available.

Most people find two of these rows land. Where two land, they usually belong to different parts of life, work in one and family in the other, and they want different responses.

Why Does Going Quiet Cost More Than the Miss Itself?

Because silence is not neutral, and the other person will fill it with something less flattering than what actually happened.

Ferrin and colleagues compared three responses to a trust violation in 2007: apology, denial, and saying nothing. Reticence performed poorly against both of the others. It left the other person's suspicion exactly where it had been, delivering neither the reassurance of an apology nor the counter-claim of a denial. Silence gets read as content, and the content it usually gets read as is that you either did not notice or did not mind.

There is a practical asymmetry underneath this that is easy to overlook. What the other person lost went beyond the call itself. They lost the ability to plan around you. Told on Tuesday, they still have Thursday and can do something else with it. Told on Sunday, or not told at all, they have already spent it waiting. The early version costs you one uncomfortable sentence and returns their week to them. The late version costs you a longer apology and returns nothing.

And the person is rarely evaluating your capacity. They are evaluating whether you know what happened. Somebody who says on Tuesday that they will not manage the call this week is legible. Somebody who reappears the following Wednesday with a full account of how busy things were is asking to be assessed on their week, which is not the question anybody was asking.

Where Does This Hide When You Would Not Call It a Promise?

In most of what functions as a commitment without ever carrying the word, which is where the majority of the volume sits.

I will look at it tonight. Send it over and I will get back to you. Let me check and I will let you know. We should do this more often. Yes, definitely, let us sort something out. Nodding while somebody describes a plan that involves you. Not objecting when a date is proposed, which everybody present correctly reads as agreement. Saying you will think about it when what you have actually done is close a subject.

Two of these deserve separate mention. The first is the reply, which is the single most commonly broken commitment there is and the one least likely to be described as a promise. Getting back to somebody is a promise. It is generally the smallest one available and the one that erodes most steadily, because there is no moment at which it visibly fails.

The second is promises to yourself, which for many people are the ones that have been broken longest and hurt most. The mechanism is identical. The version of you who decides at eleven on Sunday night to get up at six has no access to the version of you standing in the dark at six. The difference is that nobody else is there to notice, so the evidence accumulates entirely in private and gets read entirely as character.

Why Has Deciding to Be More Careful Never Held?

Because the decision gets made in one state and the promise gets made in another, and the first has no way to reach the second.

Picture the resolution honestly. It arrives on a Sunday evening, alone, calm, with the whole miserable week visible behind you. In that state the correct behavior is entirely obvious and the resolve is completely genuine. Then on Wednesday somebody is upset in front of you and the resolution is nowhere in the room, because it was formed by a person who is not currently present and who had no idea what this specific moment would feel like.

There is a second problem, and it is close to a joke. Resolving to be more careful with promises is itself a promise. It is made by the same machinery, in the same conditions, with the same warmth and the same certainty and the same lack of a slot in any particular week. Whatever explanation you accept for the first kind applies unchanged to the second.

This is why the repairs that hold are structural and slightly boring, and why the ones that fail are motivational. Nothing on this page suggests wanting it more. What is worth changing is the shape of the sentence, so that a promise made in a hot state carries something that survives into a cold one: a time, a smaller unit, or a visible checkpoint before the deadline. That is a change to the instrument rather than to the person holding it.

It is also worth saying plainly that it is not as simple as somebody telling you to just think before you speak. There is more underneath most of these than the visible half, which is why the five roots come before any of the practical material rather than after it.

What Does a Promise You Can Keep Sound Like Out Loud?

It has three properties, and none of them involve promising less.

It names a time. Not this week. Thursday evening. The time can be wrong. A wrong time still converts a promise with no location into a promise with a location, and the location is what generates the small useful pressure that gets things done. A promise with a time in it also fails visibly, on Thursday, while there is still a week left to fix it, instead of invisibly on Sunday when there is not.

It names a smaller unit than the whole thing. Not I will read your manuscript. I will read the first thirty pages and tell you by Friday what I think of them. The smaller unit is a complete deliverable rather than a reduced one, which is what separates it from simply promising less. The person gets something real on a known date instead of something large on an unknown one.

It is checkable without asking you. The best commitments produce evidence on their own. A calendar invitation sent while you are still on the phone. A message with the date in it. A shared document. Anything that means the other person does not have to decide whether it would be rude to follow up, which is a cost they are currently paying that you cannot see.

And an honest limit, because the section would otherwise be dishonest. None of this improves your forecasting. The person who has to make the call on Thursday is exactly as unreachable as before. What changes is that the failure now has a date attached to it, which means you find out on Tuesday instead of on Sunday, and Tuesday is where the useful hour lives.

Those three properties are not an invention of this page: Gollwitzer set the general form out as implementation intentions in a 1999 paper in American Psychologist, describing what changes when an intention is specified in advance as a when, a where and a how instead of as an aim, which is what a named Thursday evening does to a promise made about this week. The limit is the familiar one, that the work was built on goals people set for themselves rather than on promises made to somebody else, and a promise whose only function was to end a difficult moment does not produce a plan of any kind, which is why this section comes after the five roots rather than instead of them.

How Do You Repair One You Already Missed?

Four parts, in order, and the order does most of the work.

Name it before they raise it. Whoever names it first sets what kind of event it was. If they raise it, it is a complaint and you are responding to an accusation. If you raise it, it is an account, and it is a considerably easier conversation for both of you.

Say what happened in one sentence, without building a story around it. The week got away from me and I did not call. That is the whole thing. The instinct to explain in detail comes from wanting to be understood, and the effect is the opposite, because a long explanation asks the other person to assess whether your reasons were good enough, which puts them in a position nobody enjoys occupying.

Do not attach a new promise you have not checked. This is where most repairs go wrong, and there is a real tension in the research here worth naming. Schweitzer, Hershey and Bradlow found that a promise to change behavior sped up trust recovery, which is a genuine argument for making one. The same work found that deception is the thing that does not recover. A repair promise made on the spot, in the warm state that repair conversations produce, is generated by exactly the mechanism this page has been describing, and missing it is far more expensive than the original.

Offer the smallest thing you are certain of. Not I will definitely call this week. Ten minutes now, if they have it, or a specific time you have already looked at. Small and delivered beats large and hoped for by a margin most people badly underestimate.

What Changes for the Person Who Was Waiting?

What they lost has less to do with the call than with being able to plan around you, and that is the thing that comes back first when anything changes.

Consider what an unreliable commitment does to somebody's week. They cannot count it, so they have to hold two versions of Thursday, one where you call and one where you do not. Holding two versions of something is genuinely tiring and it is invisible to everybody except the person doing it. Over years, most people stop doing it. They collapse to the single version where you do not, which is the moment your promise stops being load-bearing in their life, and it is usually the moment they also stop mentioning it.

So the first thing that returns when this shifts has nothing to do with warmth or with forgiveness. They start putting you into plans as a fixed item again. That change is small, unannounced, and easy to miss, and it is the actual measure of whether anything has moved.

The second thing that returns is that they begin to ask you for things again. People stop asking long before they stop liking you, and the asking is a better indicator than anything either of you would say out loud.

Neither of those responds to a good conversation about it. Both respond to a sequence of small kept commitments with dates on them, which is unglamorous, and which works.

Where Does This Thinking Come From?

The hot-cold empathy gap entered the literature in 1996 in a paper on visceral influences on behavior. Read and van Leeuwen published the snack-choice studies in 1998 in Organizational Behavior and Human Decision Processes, finding that current hunger shaped choices made for a week later. Sayette and colleagues published the cold-to-hot work with smokers in 2008 in Psychological Science. Badger and colleagues reported a related finding in 2007 in the Journal of Health Economics, showing that current craving substantially changed how people valued a future dose.

On the distance between intending and doing, Armitage and Conner published their meta-analysis of 185 studies in 2001 in the British Journal of Social Psychology, reporting that intention and perceived control accounted for roughly a quarter of the variance in behavior. Sniehotta, Scholz and Schwarzer published the bridging work in 2005 in Psychology and Health, finding that planning and action control carried most of the remaining distance.

On the future self, Pronin, Olivola and Kennedy published the allocation studies in 2008 in Personality and Social Psychology Bulletin. Bartels and Urminsky published the identity-continuity work in 2011 in the Journal of Consumer Research. Hershfield and colleagues published the neural work on future-self representation in 2009 in Social Cognitive and Affective Neuroscience; the applied interventions built on that line have a weaker record and are described here with that caveat.

Herrnstein and Prelec described melioration in 1991 in the Journal of Economic Perspectives, showing how a series of locally sound choices can produce an aggregate nobody would have chosen. On trust, Schweitzer, Hershey and Bradlow published the promises-and-lies experiments in 2006 in Organizational Behavior and Human Decision Processes, and Ferrin and colleagues published the comparison of apology, denial and silence in 2007 in the Journal of Applied Psychology.