Users don't lie in interviews. They guess.

You show someone the thing you’ve been building. They lean in, nod, and say the sentence you’ve been waiting for: “Oh, I’d definitely use this.” You write it down. Three months later, you ship, and they don’t.

It’s tempting to file this under dishonesty. It isn’t. That person wasn’t lying to you. They were doing something much harder and much more forgivable: predicting the behavior of a future self they’ve never met, under conditions that don’t exist yet, while sitting across from a hopeful founder they’d rather not disappoint. Almost nobody is good at that. The “lie” in “users lie in interviews” is really three ordinary failures of self-report, and once you can name them, you can ask around them.

The first is that we’re bad narrators of our own minds. People will give you a confident, articulate reason for a preference that had nothing to do with the reason. This isn’t a character flaw; it’s how introspection works. We reach for a plausible story and report that, mistaking it for the actual cause. So when you ask “why would you use this?”, you don’t get the mechanism, you get a well-formed guess dressed as an explanation.

The second is that an opinion about the future is free, and everyone can afford to be generous. “Would you use a tool that does X?” costs the person nothing to answer yes. There’s no calendar, no budget, no switching cost in the room, just a hypothetical and a friendly face. Hypotheticals are where people are their most optimistic and least accurate, because the version of them that would actually adopt your product has to overcome friction that the question conveniently deleted.

The third is the desire not to hurt you. You built this. You’re excited. Most people, most of the time, will round their honest “meh” up to a kind “that’s cool” rather than watch your face fall. The warmer the room, the worse the data.

The move that fixes all three at once is to stop asking about the future and the abstract, and start asking about the past and the specific. You can’t verify an opinion. You can absolutely verify a story. So instead of “would you use a tool that organizes your receipts?”, you ask: “Tell me about the last time you had to find a receipt. What were you doing? What did you try? How long did it take?” Now the person isn’t forecasting, they’re remembering. And memory, while imperfect, leaves evidence: they either did or didn’t already build a spreadsheet, already pay for something, already hack a workaround at 11pm. Real problems have a trail of scar tissue. Fake ones don’t.

This is the whole spirit of The Mom Test — the rule that you should be able to ask your questions of your own mom and still get useful data, because the questions are about her life, not about your idea. If a question can only be answered “yes, that’s a great idea,” it’s a bad question. If the honest answer might be “actually I’ve never bothered,” you’re finally learning something.

A few concrete swaps:

Instead of “Would you pay for this?” ask “What are you paying for today to deal with this, and what does it cost you?” Money already spent is a fact. Money hypothetically offered is a compliment.

Instead of “How often would you use it?” ask “How many times did you do this in the last week?” Frequency they’ll actually give you a number for is frequency you can trust; frequency they’d project is fiction.

Instead of “Do you like this?” ask nothing, and watch. Hand them the thing and shut up. What they do with their hands in the next sixty seconds is worth more than what they say in the next five minutes.

The reframe that changes how you run every conversation from here: the goal of a user interview isn’t to get someone to endorse your idea. It’s to find out what they already do when you’re not in the room, because that, and not their kind forecast of a stranger, is what they’ll keep doing after you ship. Users aren’t lying to you. They’re guessing about a future they can’t see. Your job is to stop asking them to guess.