You Do Not Know Why You Bought It
People asked why they chose something construct a plausible reason rather than retrieving the real one. The 1977 paper that showed this also says exactly when the answer can be trusted, and that half gets dropped.
Key takeaways
- Nisbett and Wilson reviewed the evidence in 1977 and concluded there may be little or no direct introspective access to higher order cognitive processes.
- People are sometimes unaware of the stimulus that influenced a response, sometimes unaware of the response, and sometimes unaware that the stimulus affected it.
- Reports on those processes are drawn from implicit causal theories about what plausibly causes what, rather than from introspection.
- The boundary condition is the part usually dropped: accurate reports do occur, where the influential stimulus is salient and is a plausible cause of the response.
- So a direct question about an obvious attribute is a reasonable instrument, and a direct question about why one option beat another is not.
Ask yourself why you picked the wine you picked last time.
Whatever answer arrives, you did not retrieve it. You built it. A self-report about your own motives is a weaker instrument than it feels, and the rules governing exactly how weak have been known since 1977.
Our position, up front, since it governs how the rest reads. The strong version of this claim is overstated and the weak version is devastating for an entire industry. People are not sealed off from their own minds. They are unreliable about causes specifically, the unreliability is predictable, and predictable means workable.
What did Nisbett and Wilson actually claim?
Kevin Simler and Robin Hanson's The Elephant in the Brain is where this site came to the problem: their argument is that people routinely act on motives they do not report and may not have access to, and that the gap between the real reason and the stated one is systematic rather than random. Their case is built from evolutionary reasoning and examples rather than from experiment, which is why the paper underneath it matters more.
Richard Nisbett and Timothy Wilson, in Psychological Review in 1977, reviewed the evidence on whether people can report on their own mental processes. It is one of the most cited papers in psychology and the summary it gets reduced to is cruder than what it says.
Their finding has three parts. People are sometimes unaware of a stimulus that importantly influenced a response. Sometimes unaware of the response itself. And sometimes unaware that the stimulus affected the response at all.
Their explanation is the part that matters here. When people report on the processes connecting a stimulus to their response, they are not introspecting. They are drawing on implicit causal theories, judgements about whether a given thing is a plausible cause of a given behaviour.
So a shopper asked why they chose a product does not retrieve the cause. They generate the most plausible-sounding candidate from a theory they already hold about what moves people like them.
### The three separate failures
It is worth separating the three, because they fail in different ways and only one of them is about introspection.
Ask about something obvious and visible and the answer is probably fine. Ask about something subtle and the answer is a theory.
The first is being unaware of the stimulus. Something in the environment moved you and you did not register it. That is an attention failure and it is the easiest to accept.
The second is being unaware of the response. You behaved differently and did not notice that you had. Harder to accept and well documented.
The third is the one the paper is really about. You registered the stimulus, you registered the response, and you are wrong about whether one caused the other. That is not a gap in perception. It is a theory about yourself, applied confidently to a case it does not fit.
When can a self-report be trusted?
Here is the reframe, and it comes from the paper rather than from a critic.
Nisbett and Wilson are explicit that accurate reports do occur, and they say when. Reports will be accurate where the influential stimulus is salient and is a plausible cause of the response. They will be inaccurate where the stimulus is not salient, or where it is not a plausible cause.
That is a usable rule, and it turns a famous pessimism into a prediction.
### Why does price work and position not?
Price is salient, you looked at it, and it is a plausible cause of a purchase in anybody's theory of shopping. So a self-report naming price is likely to be roughly right.
Shelf position is not salient, you did not attend to it, and almost nobody holds a theory in which the middle shelf causes purchases. So a self-report will not name it, whether or not it was decisive. The absence of position from the answer is not evidence that position did not matter.
> Ask about something obvious and visible and the answer is probably fine. Ask about something subtle and the answer is a theory.
So "why did you buy this coat" will often get a serviceable answer, because price and fit and colour are salient and plausible. "Why did you choose the second-cheapest wine" will not, because the position on the list is neither.
What does asking customers why return?
The consequence is specific and it is not that asking people things is useless.
A self-report about a salient attribute is a reasonable instrument. A direct question about why one option was preferred over another, where the deciding factor was layout, order, a default, a colour or a price ending, is an instrument that will reliably return a confident and wrong answer. The respondent is not lying. The self-report is doing exactly what the paper says it does.
Our own piece on decoy options makes the same point from the other direction, and this is also where The Elephant in the Brain is both useful and loose. Simler and Hanson are right that stated reasons systematically miss real ones, and they do not supply the condition under which a stated reason is reliable. Nisbett and Wilson do, and it is the difference between an argument and an instrument.
The same gap explains a pattern anybody who has read focus-group output will recognise. The reasons given are always sensible, always in the register of value for money and quality, and almost never mention the arrangement of the shelf. Our own piece on decoy options makes the same point from the other direction.
The methodological escape is not cleverer questions. It is measuring behaviour under manipulation rather than asking for causes, which is what every study this publication relies on actually does.
### How should you grade a study?
It also gives you a way to grade a study in one question. Did the researchers change something and watch what people did, or did they ask people what they would do and why?
The first design can detect a cause the participant never noticed. The second cannot, by construction, and no sample size fixes it. That is the difference between the studies we cite and most of what reaches a reader as consumer research.
### Where the strong version goes wrong
Two limits, and the first is on us as much as on anybody.
This is a 1977 review, not an experiment, and reviews summarise the state of a literature at a moment. Much has happened since in work on introspection, metacognition and confabulation, and we have not surveyed it. So what we are reporting is the paper's claim and its own stated boundary, not the current consensus.
And the paper is routinely stretched into the claim that people have no self-knowledge at all, which it explicitly denies. Its own boundary condition, salient and plausible stimuli yielding accurate reports, is the thing most summaries drop, and we have dropped it ourselves in earlier pieces when citing self-report limitations in passing.
So why did you pick that wine?
So the question at the top of this article has a two-part answer, which is the honest place to end.
If you picked the wine because you recognised the grape, or because of the price, you probably did retrieve that and your account is fine. If you picked it from a ranked list, then what a ranking is actually recording is a separate problem again. If you picked it because it was second from the bottom of the list, or because of where it sat on the shelf, your account will not contain that and will sound just as convincing.
The difficulty is that the feeling is identical in both cases. No internal signal distinguishes a retrieved reason from a manufactured one, which is why a confident self-report is not evidence of an accurate one, which is why the useful move is not to trust or distrust your own account but to ask what else was on the screen at the time.
Common questions
Can people explain why they bought something?
Sometimes. Nisbett and Wilson's 1977 review concluded there may be little or no direct introspective access to the processes linking a stimulus to a response, and that reports on those processes are generated from implicit theories about plausible causes. They also specified the exception: reports are accurate where the influential stimulus was salient and is a plausible cause of the response.
Is a self-report always unreliable?
No, and the paper explicitly denies the blanket version. Ask about an obvious and plausible factor, such as price or fit, and the answer is likely to be roughly right. Ask why one option beat another where the deciding factor was layout, order, a default or a price ending, and the answer will be confident and wrong.
Why do focus groups give misleading answers?
Because they ask for causes. Respondents are not lying; they produce the most plausible candidate from a theory about what moves people, and shelf arrangement does not appear in that theory. The reasons given will be sensible, in the register of quality and value, and silent about the things the research shows are decisive.
What should be used instead of asking why?
Measuring behaviour under manipulation. Change one thing, hold the rest constant, and watch what people do. That design can detect a cause the participant never noticed, which is precisely what a self-report cannot do.
Sources
Nisbett, R. E., & Wilson, T. D. (1977). 'Telling more than we can know: verbal reports on mental processes.' Psychological Review, 84(3), 231-259. Reviews evidence suggesting little or no direct introspective access to higher order cognitive processes; subjects are sometimes unaware of an influential stimulus, of the response, or that the stimulus affected the response; reports are based on a priori implicit causal theories about whether a stimulus is a plausible cause of a response; accurate reports will occur where influential stimuli are salient and plausible causes, and will not where they are not. Verified at the published record 2026-10-05. doi:10.1037/0033-295x.84.3.231
Simler, K., & Hanson, R. (2018). The Elephant in the Brain. Oxford University Press. The hidden-motives argument. Read in full in this vault, 2026-09-14.
A note on what this is: a review rather than an experiment, published in 1977, summarising a literature at that moment. Work on introspection, metacognition and confabulation has continued and this article has not surveyed it, so what is reported is the paper's claim and its own stated boundary rather than the present consensus.