01Two kinds of preference
Stated preference is what someone tells you when you ask — on a survey, in an interview, in a “would you use this?” poll. Revealed preference is what they do when no one is asking: what they buy, what they complain about, what they quietly build a workaround for at 11pm.
The two diverge constantly, and predictably. Stated preference is cheap to give and socially shaped — people answer the question they think you want answered. Revealed preference costs something real, so it can’t be faked.
02Why surveys flatter
A survey puts your idea in front of someone and asks them to imagine a future. Imagining is free, and most people are kind, so the answers skew positive — the classic “sure, I’d use that” that never converts to a signup.
It gets worse with price. Asked what they’d pay, respondents anchor on being agreeable rather than on their actual budget. This is why an idea can survey beautifully and still die on contact with a real checkout: you measured stated preference and shipped to revealed.
03Why unprompted forum talk is closer to the truth
When someone posts on Reddit that they’ve wasted another evening fighting a problem, nobody prompted them. There’s no product in front of them to be polite about, no researcher to please — just a real frustration worth the effort of writing up.
That’s as close to revealed preference as text gets: people describing what they actually did, paid for, and abandoned, in their own words, because they wanted to — not because a survey made them.
04What revealed preference looks like in a thread
The tells that someone is describing behaviour, not just an opinion:
- A workaround they already built or bought, not a feature they’d “like”
- A specific price they pay or left — a number with a decision behind it
- Time spent — “an hour every morning”, not “it’s a bit annoying”
- A tool named and judged from use, not from a landing page
- A switch they actually made, with the trigger that caused it
The honest caveat
Reddit isn’t pure revealed preference — it’s what people choose to say, so it carries its own selection bias (the frustrated post more than the satisfied). It’s closer to behaviour than a survey, not a substitute for a live test. Treat it as the strongest cheap signal, then confirm with a real checkout.
Compare the methods directly
Surveys, interviews, and forum mining each capture a different slice — here’s when to reach for which.
Reddit vs surveys vs user interviews →05Turning the distinction into a decision
Discount the enthusiasm
Strip out the “sounds great”, the upvotes on a wishlist post, the “I’d totally pay for that”. None of it cost anyone anything, so treat it as noise until proven otherwise — not as demand.
Weight the costly signals
A workaround someone built, a number they actually paid or churned over, an hour lost every morning, a tool switched away from — each of these carries a decision behind it. Rank by how much the behaviour cost the person, not by how loudly it was stated.
Cluster the strongest revealed signals
Look for the same costly behaviour described by different people in different threads. One person is an anecdote; the same workaround appearing five times is a pattern worth betting on.
Confirm the top pattern with a real test
Revealed talk narrows the field cheaply, but it is still talk. Put the strongest cluster in front of a real checkout, waitlist, or pre-order and watch what people do. A ranked, sourced report of real threads tells you where to point that test — it does not replace it.
Frequently asked questions
Stated preference is what someone tells you when asked — on a survey, in an interview, in a poll. Revealed preference is what they do when no one is asking: what they buy, abandon, complain about, or build a workaround for. Stated preference is cheap and socially shaped; revealed preference costs something real, so it is far harder to fake.
Surveys ask people to imagine a future, and imagining is free. Most respondents are agreeable, so answers skew positive — the “sure, I’d use that” that never converts. There is no cost to saying yes and a mild social reward for it, so you measure goodwill rather than intent. The gap shows up the moment a real checkout asks for money.
Rarely. Asked a price in the abstract, people anchor on sounding reasonable rather than on their real budget, so quoted figures drift from what they would actually spend. A number someone already pays — or quit over — is worth far more than a hypothetical one. To get a trustworthy price, watch a real transaction, not an answer to a pricing question.
Look for behaviour that cost something: money spent, time lost, a workaround built, a tool adopted or dropped. Online, unprompted forum posts come close — people describing what they actually did in their own words, because they wanted to. The cleanest measure is still a live test, like a checkout or pre-order, where the action itself proves the preference.