What People Remember Is Not What Happened

In a randomised trial of 682 patients undergoing colonoscopy, one group had a short, less uncomfortable interval added to the end of the procedure. That group experienced more total discomfort and remembered the procedure as less unpleasant — and was somewhat more likely to return for follow-up.¹

That is the cleanest available demonstration of two things: that remembered experience and lived experience are different quantities, and that the difference is engineerable.

The mechanism

Redelmeier and Kahneman had already shown the pattern with real-time pain recording in colonoscopy and lithotripsy patients: total remembered pain correlated with the peak intensity and with the intensity of the final minutes, while the duration of the procedure barely registered — across procedures ranging from four to sixty-nine minutes.²

The earlier laboratory version is starker. In the cold-pressor experiment, participants preferred to repeat a longer trial — sixty seconds of cold water plus thirty more seconds as it warmed slightly — over a shorter one, despite more total discomfort, because it ended better.³

Two names for the two halves: the peak-end rule is that memory is dominated by the most intense moment and the final moment. Duration neglect is that how long it lasted contributes almost nothing.

Does it hold up?

This desk has spent several articles on findings that did not survive scrutiny, so it is worth being specific about one that did. A 2022 meta-analysis pooling 174 effect sizes found a combined peak-end effect on retrospective evaluations of r = .581, described as large and robust across boundary conditions, with the duration effect essentially nil.⁴

And there is a quiet caveat inside that same meta-analysis which almost nobody quotes: the peak-end composite predicted memory only comparably to a simple average of the whole experience. Averaging every moment worked nearly as well.

The strong version — that only the peak and the ending matter — is not what the evidence supports. The defensible version is that the peak and the ending matter far more than duration.

That distinction is the difference between designing a good ending and neglecting the middle.

What it means for delivering anything

Four applications, ordered by how cheap they are:

1. Do not let the engagement end on the invoice. For most service businesses, the final interaction a client has is a bill or a chase for payment. That is the moment the memory is built from. A short closing conversation — what was delivered, what to watch for, what comes next — costs fifteen minutes and occupies the position in memory that the invoice otherwise takes.
2. Front-load the difficulty. If a project has an unpleasant phase — data migration, a switchover, a hard conversation about scope — schedule it early rather than late. The same difficulty is remembered differently depending on where it sits.
3. Build one peak deliberately. A single moment that exceeds expectation contributes more to the memory than a uniformly slightly-better experience. This is not the same as being better throughout, and it is considerably cheaper.
4. Stop optimising for speed alone. Duration neglect cuts both ways: a faster delivery that ends badly is remembered worse than a slower one that ends well. Speed is worth less than you think and endings are worth more.

Where it does not apply

The rule describes memory of a single bounded episode. It does not rescue a product that inflicts a fresh bad memory every day — recurring experiences are evaluated differently, and no amount of good endings compensates for a tool that is unpleasant every Tuesday.

Duration neglect is also strongest for short, aversive experiences of the clinical kind. For long pleasant ones, duration accumulates at least some remembered value.

The ethical line, since it is a real one

This research can be read as instructions for manipulating memory, and it would be dishonest not to say so. The line we would draw: it is legitimate to design where difficulty sits and how an engagement closes. It is not legitimate to use a good ending to obscure a bad delivery, because the client's memory being favourable does not make your work good — it makes their next decision worse informed.

Used correctly, this is not a trick. It is the recognition that finishing well is a real part of the work, and that most operations treat the ending as administrative overhead rather than as the thing the client will remember.