Ask About Yesterday: How to Interview

"Would you use an app that tracks your expenses automatically?"

"Oh, definitely. I've been meaning to sort my finances out for ages."

Six people say versions of this. You build it. Twelve people download it, four open it twice, nobody stays. And the honest post-mortem is not that the six lied — it's that you asked a question whose answer could only ever be yes.

Why people mislead you without lying

Three mechanisms, none of them dishonest.

Politeness. You're in the room, visibly invested, and they're a decent person. Criticising your idea to your face costs them something and gains them nothing. So they soften — and the softening looks exactly like validation.

Prediction is hard. Asking "would you use this?" asks someone to simulate a future version of themselves in circumstances that don't exist yet. People are poor at this about their own behaviour, and there's no way to be good at it.

We invent reasons. Richard Nisbett and Timothy Wilson showed in 1977 that people confidently report the reasons for their choices even when those reasons demonstrably weren't the cause. Ask "why did you do that?" and you'll get a fluent, plausible, sincerely believed answer that may have nothing to do with what happened.

Put together: the further a question is from something that actually happened, the less the answer is worth.

Don't ask what they would do. Ask what they did.

The Mom Test

Rob Fitzpatrick named the standard for this: a good question is one your mother couldn't give you a falsely encouraging answer to, however much she loves you. "Do you think my idea is good?" fails. "Walk me through the last time you tried to do this" passes — she'd have to invent an entire fictional Tuesday to flatter you.

Three rules follow, and they're the whole method:

Talk about their life, not your idea. Your idea does not need to come up. If it does, it's at the end.
Ask about specifics in the past, not generalities about the future. "Last time," not "usually" or "would."
Listen more than you talk. If you're speaking more than a quarter of the time, you're conducting a demo, not an interview.

Three bad questions and what to ask instead

"Would you use X?" → "Tell me about the last time you needed to do this. What did you actually do?"

The replacement gets you a real event with real details: the tool they used, how long it took, what annoyed them, and whether they finished. All of it checkable, none of it invented.

"Do you like this design?" → "What would you do here? What do you expect to happen if you press that?"

The first asks for a compliment. The second asks for a prediction you can compare against reality — and the gap between them is a usability finding.

"How often do you do this?" → "When was the last time? And before that?"

Frequency estimates are terrible. Two dated events give you a real interval, and the difference between "a few times a week" and "twice since spring" changes what you build.

Two more to retire: "Would you pay for it?" (everyone says yes; ask what they currently pay for and how the decision was made) and "What features do you need?" (you're asking for a solution; ask what they did when the thing they needed didn't exist).

The ladder and the well

Two digging techniques, in different directions.

Laddering goes upward, from a stated preference to the value underneath it. "Why is that important to you?" — repeated, gently, three or four times. Someone says they want the export button on the main screen. Why? To send it to their boss on Fridays. Why does that matter? Because otherwise they get chased on Monday morning. And now you know that the real job is being visibly done by Friday, which might not need an export button at all.

The five whys goes downward, from a symptom to a cause. They abandoned checkout. Why? Didn't have the card. Why? It's the company card, kept by someone else. Why does that matter here? Because approval takes two days and the session expires in twenty minutes. You have just discovered that your checkout was designed for a person who doesn't exist in this company.

Use them with a light hand — four consecutive "why"s starts to feel like an interrogation. Alternate with "tell me more about that" and "what happened next."

The most underused tool is silence

Ask your question and then stop. Count three seconds after they seem to have finished. Almost always they keep going, and the second half is the useful half — the first was the prepared answer, the second is the actual one.

Related discipline: never defend your design during an interview. The moment you explain how it's supposed to work, you have taught them the answer and destroyed the sample. If they're stuck, "what would you try?" or "what were you expecting?" — and then silence again. It will feel rude. It isn't.

Look for the cost they already paid

Words are cheap; effort isn't. The strongest signals in any interview are the things a person has already spent something on:

A workaround. A spreadsheet, a notebook, a naming convention, a second account. A workaround is a specification written by a user in their own hand.
Money. They already pay for something adjacent, or they paid once and stopped.
Time. They do a boring thing manually every week and have never questioned it.
Complaint. They've told someone, asked for it, or shouted about it publicly.
Search. They went looking for a solution — that means the problem crossed the threshold where they took action.

Fitzpatrick's term for the equivalent in the other direction is commitment: enthusiasm is worthless, but a person who gives you their time, their money or an introduction to their boss has told you something true.

The shape of a session

Forty-five minutes is plenty. Roughly:

5 min — warm-up. Explain there are no right answers, you didn't design this, and you want them to be blunt. Ask permission to record.
10 min — context. Their role, their week, where this task sits among everything else they do.
20 min — the story. One concrete recent occasion, walked through end to end, with digging at every point of friction.
10 min — the rest. Other occasions, and only now, if it's relevant, the thing you're building — shown, not pitched.

How many people? Five to eight per distinct segment. You stop when new conversations stop producing new information — saturation — which usually arrives sooner than people expect.

Who? Recruit by behaviour, not demographics: "people who did X in the last month." And avoid an all-friends sample; friends are the politest respondents on earth.

Write it up within a day. Memory rewrites interviews to agree with whatever you concluded afterwards. Raw quotes, in their words, before your interpretation gets on top of them.

In practice

Rewrite your questions before every session. Take each one and check: could this be answered about something that already happened? If not, rewrite it until it can.

Record the words, not the summary. Verbatim quotes are the currency of research. "Users found it confusing" convinces nobody; "I thought that button would delete my whole file" ends the argument.

Count your talking. Record a session and check how much of it is you. Above 25% and the interview was mostly about you.

Try three seconds of silence today. In any conversation, not just research. It's a startling amount of extra information for zero effort.

Ask "what did you do instead?" whenever someone describes a problem. That question finds workarounds, and workarounds are the highest-value thing in the room.

Interview one person about a product you didn't make. No stake, no defensiveness — the fastest way to practise the technique before it matters.

Check yourself

Close the article and answer in your own words:

Name the three reasons people give you unreliable answers without lying.
What is the Mom Test, and how do you tell whether a question passes it?
Rewrite "would you use this?" and "do you like this design?" into good questions.
What's the difference between laddering and the five whys, and when do you use each?
Why must you not explain your design during an interview?
Why is a workaround worth more than an enthusiastic opinion?
How do you know when you've interviewed enough people?

In short

People mislead you out of politeness, poor self-prediction and invented reasons — not dishonesty. The further from a real event, the worse the answer.
The Mom Test: ask what even your mother couldn't answer flatteringly. Talk about their life, ask about the past, listen more than you speak.
Replace "would you," "do you like" and "how often" with "the last time," "what would you do here" and "when exactly."
Laddering climbs from preference to underlying goal; the five whys digs from symptom to cause.
Silence for three seconds after the answer produces the real answer. Never defend your design mid-session.
Trust what people have already spent: workarounds, money, time, complaints, searches. A workaround is a spec in the user's handwriting.
Five to eight people per segment, recruited by behaviour, written up the same day in their own words.