People don't remember experiences as averages. They remember them by their peak — the most intense moment — and their ending. Everything in between matters far less than you think.
In the late 1980s, psychologist Daniel Kahneman and colleagues ran a series of experiments that produced one of the most counterintuitive findings in all of behavioural science. They gave participants a colonoscopy — deliberately varying whether the procedure ended abruptly at peak pain or with a longer but less painful final period — and then asked them to rate the overall experience.
The participants whose procedure was extended with a gentler ending remembered the entire experience as less painful, even though they had objectively experienced more total pain and duration. The memory of the experience was dominated almost entirely by two moments: the most intense point, and the ending. Everything in between was largely irretrievable.
This is the Peak-End Rule: we evaluate experiences retrospectively based on a mental snapshot of how they felt at their worst (or best) and how they concluded — not on any kind of integrated average. For designers, this means that two moments matter more than all the others combined: the most emotionally loaded interaction in your product, and the final moment of any experience.
Kahneman ran participants through a series of experiments involving deliberately uncomfortable tasks. The most famous involved immersing one hand in 14°C water for 60 seconds (unpleasant) versus immersing the same hand for 60 seconds at 14°C followed by 30 additional seconds where the temperature was secretly raised to 15°C (slightly less unpleasant). When asked which trial they would repeat, participants overwhelmingly chose the longer trial — the one with objectively more total discomfort — because it ended better.
The mechanism is what Kahneman calls “duration neglect”: our retrospective evaluation of experiences ignores their length almost entirely. What gets encoded into memory is a rough average of the peak moment and the final moment. This is not a failure of reasoning — it is how the remembering self fundamentally works. And the remembering self is the one that decides whether to come back.
“The remembering self is the one that keeps score, and dominates our choices.”
— Daniel Kahneman
The moment a user completes onboarding is one of the most important moments in your entire product. It's the end of their first significant experience, and per the Peak-End Rule, it will anchor their memory of everything that came before it. A cold, transactional completion screen communicates that the product views onboarding as a chore to get through. A warm, celebratory one signals that the product values the user's time and is genuinely glad they're here.
Checkbox, two lines, a button. Functional. Forgettable. The experience ends with no emotional signal.
Warm tone, clear next step, a moment of delight. The experience ends with a feeling worth remembering.
Notice also that the good screen includes a suggested first step. This is a Peak-End design principle applied to the beginning of the next experience: it ensures the next moment users encounter is also a positive one, extending the emotional arc rather than dropping users into an empty state with no guidance.
If there is a single moment in any web product most likely to become the “peak” of a negative experience, it's the error page. Something has gone wrong. The user is lost, frustrated, or confused. This is exactly the moment the Peak-End Rule predicts will disproportionately colour their memory of the entire visit.
A cold error page — a large “404”, a technical error message, a stack trace — confirms every doubt. A warm error page acknowledges the frustration, speaks like a human, and immediately offers concrete paths back to something useful.
A stack trace on a public page. The peak negative moment ends with confusion and no clear path forward.
Same error — handled with warmth, clear explanation, and three immediate recovery paths.
The warm error page doesn't minimise the error — it acknowledges it plainly. “This page doesn't exist” is honest. What changes is everything around that honest acknowledgment: the tone, the offered paths forward, and the absence of technical noise that the user cannot act on.
Every AI response has an ending. In most current implementations, that ending is some variation of “Let me know if you have any other questions” — a closing phrase that has become so universal it is now essentially invisible. It communicates nothing, offers nothing, and leaves the user in exactly the same position they were in before the last sentence.
Per the Peak-End Rule, the final line of an AI response is disproportionately likely to be what the user remembers about that response. An AI that ends every answer with a generic filler phrase trains users to stop reading before the end. An AI that ends with a concrete next step creates a closing moment that is actually useful.
The response's final moment adds nothing. Users learn to tune out the last line of every AI answer.
The last moment of the response is the most useful one — a concrete offer that turns the conversation into a working session.
The content of the main response is identical on both screens. The difference is entirely in what happens last. The action card ending creates a moment of genuine forward momentum — users don't just have information, they have an obvious next move. This is Peak-End thinking applied to conversational AI: engineer the ending of every significant exchange to be the most useful, most memorable moment of it.
The Peak-End Rule has a specific implication that is easy to test: identify the two or three most emotionally loaded moments in your product — the moments where something succeeds or fails most visibly — and the endpoints of your most common user journeys. These are the moments that write your product's memory. Everything that happens in between matters considerably less than you might assume.
Error states, completion screens, empty states, and the final line of an AI response are all high-leverage targets precisely because they are often under-designed. Teams invest in the onboarding flow but not the onboarding completion. They invest in the feature but not the success state. They invest in the AI model but not how its responses end. The Peak-End Rule tells you exactly where that investment would compound the most.
Kahneman, D., Fredrickson, B. L., Schreiber, C. A., & Redelmeier, D. A. (1993). When more pain is preferred to less: Adding a better end. Psychological Science, 4(6), 401–405. · Redelmeier, D. A., & Kahneman, D. (1996). Patients' memories of painful medical treatments. Pain, 66(1), 3–8.