The Mom Test, Applied: AI Prompts to Interview Customers Who Won’t Lie to You

You describe your idea to ten people, nine say "I'd totally buy that," you build it, and nobody buys. That gap is the exact problem Rob Fitzpatrick's The Mom Test solves. It is the premise behind The Mom Test applied to customer interviews: feedback is almost worthless by default, because people lie to be nice, and the fix isn't better instincts. It's better questions.

The title comes from a thought experiment: ask your mom if your business is a good idea and she'll say yes, because she loves you, not because she's evaluated the market. Everyone you talk to is a little bit your mom. Below are the three rules, the questions that follow from them, and a copy-paste prompt that reads a raw transcript and strips out every compliment and wishlist, leaving the only thing that matters: what people have actually done and actually paid for.

What is the Mom Test?

The Mom Test is a set of rules for talking to customers so that even someone motivated to lie to you (your mom, a polite stranger, a friend who won't crush you) can't give you false encouragement. You do it by never mentioning your idea and only asking about their real, already-lived experience.

The Mom Test says: don't ask people whether your idea is good, ask about their life. Facts about what they've already done and paid for are reliable; opinions, compliments, and promises about the future are worthless.

The mistake almost every founder makes is pitching disguised as research. Say "would you use an app that does X?" and you've just handed the other person the answer you want, and they'll hand it right back. The move is to make the conversation about them so their real behavior does the talking, and the three rules below keep it there.

What are the three rules of the Mom Test?

Fitzpatrick boils good customer conversations down to three rules. Break any one of them and you're collecting flattery, not data.

Rule 1: Talk about their life, not your idea. The moment you describe what you're building, the interview is contaminated, because now they're reacting to you instead of reporting their reality. Ask how they currently handle the problem, what tools they cobble together, what it costs them. You learn more from "walk me through the last time this happened" than from any reaction to your pitch.

Rule 2: Ask about specifics in the past, not generics about the future. "Would you buy this?" and "do you usually…" both invite fiction, because people answer hypotheticals with their idealized self. Pin every question to a concrete past event. Not "do you struggle with bookkeeping?" but "what happened last time your books were a mess?" The past is factual. The future is a wish.

Rule 3: Talk less, listen more. If you're talking, you're not learning, and you're probably leading them. Aim to let them speak the large majority of the time. Ask a short question, then shut up and let the silence pull the real story out. Most operators fill the gap and step on the exact detail they needed.

All three move the spotlight off your idea and onto their behavior. That is the whole game.

What questions actually pass the Mom Test?

Here's the practical part: the questions that follow the three rules, written so you can run them live. Swap the bracketed problem for yours, ask them in roughly this order, and do not pitch.

• "Walk me through the last time you dealt with [the problem]. What actually happened?"
• "How do you handle [the problem] today? What tools or workarounds are you using?"
• "How much time or money does [the problem] cost you right now?" (Get a number.)
• "What have you already tried to fix it, and why did that not stick? Did you spend money on it?"
• "Where does this rank against the other fires you're fighting?"
• "Who else feels this, and who signs off on buying a fix?"

Two questions force real commitment: "Can you introduce me to someone else with this problem?" and, once you have something to show, "Would you put down a deposit today?" A referral costs social capital and a deposit costs money, so both beat any verbal "yes." Fitzpatrick's blunt version: a compliment is worthless; a commitment of time, reputation, or cash is the only thing that counts.

How do you strip the flattery out of a transcript?

This is where AI earns its seat. After the call, your transcript is full of "that's a cool idea" and "you should add feature X," none of which is data. The job is to separate three things: facts (what they've done and paid), compliments (worthless), and wishlists (worthless until backed by a commitment). By hand that's slow and biased, because you'll keep the nice parts without meaning to. A model won't.

Record every interview (with permission), get a transcript, and run the prompt below. It grades the call against the three rules, throws out the flattery, and hands back only the buying signal plus the commitments you secured.

You are a customer-development analyst trained on Rob Fitzpatrick's The Mom Test.
I will paste a raw interview transcript. My product idea is: [ONE-SENTENCE IDEA].
The customer's role/segment is: [ROLE OR SEGMENT].

Analyze the transcript against the three rules of The Mom Test:
1. Talk about their life, not the idea.
2. Ask about specifics in the past, not generics about the future.
3. Listen more than you talk.

Return exactly these sections and nothing else:

1. HARD FACTS (things they have actually DONE or PAID FOR)
   - Bullet each concrete past behavior, tool, workaround, or dollar/time cost.
   - Ignore anything hypothetical or future-tense.

2. COMPLIMENTS TO DISCARD
   - Quote every piece of flattery or "I'd use that" and mark it [NO SIGNAL].

3. WISHLIST / FEATURE REQUESTS
   - List feature asks, but label each: is it backed by a real past
     workaround, or just a wish? Wishes are [NO SIGNAL] until validated.

4. COMMITMENTS SECURED (the only real currency)
   - Did they give up time, reputation, or money? (a referral, a next
     meeting, a deposit, a signed LOI). Quote each, or write "NONE."

5. BUYING-SIGNAL VERDICT
   - Rate 1-5 how strong the evidence is that this person will PAY.
   - Base it only on Facts + Commitments, never on Compliments/Wishlist.

6. WHERE I BROKE THE RULES
   - Quote any moment I pitched, led the witness, asked about the future,
     or talked too much. One-line fix for each.

7. THREE BETTER QUESTIONS to ask next time, following the three rules.

Be blunt. If there is no real signal, say so plainly. Do not soften it.

Run it after every conversation and section 5 becomes your scoreboard: ten calls averaging a 2 means the problem isn't painful enough to sell into yet. Section 4, the commitments you actually secured, is the honest measure of whether you have a business.

Honest limits

The Mom Test sharpens the questions, but it doesn't pick who to ask. Interview the wrong segment perfectly and you get clean, reliable data about people who will never buy. It also skews toward existing behavior, weaker for a brand-new category with no "last time" to ask about. And the analyzer is only as good as the transcript: lead the witness on the call and the model faithfully reports the biased answers you fished for. AI cleans up flattery; it can't fix a rigged interview. Treat the verdict as a signal to weigh, not a green light to build.

Where to take this next

Do this before you write a line of copy or spin up a funnel, because everything downstream inherits whatever you get wrong here. Run five interviews this week using the script above, feed each transcript through the prompt, and watch how fast "everyone loves it" collapses into "two people have actually paid to solve this." Then let those facts become your positioning. The verbatim pains you collect are the raw material for copy that sells without sounding like a pitch, they tell you which fears to dissolve when you build an offer they can't refuse, and they feed the listening tools in Never Split the Difference, applied. Want your five transcripts pressure-tested by operators who've run this drill? Bring your interviews to the Asset Academy community and get the verdict double-checked.

Frequently Asked Questions

What is the Mom Test in simple terms?

It's a way to talk to customers so nobody can give you false encouragement, even if they want to. Instead of asking whether your idea is good (which invites polite lies), you ask only about their real, already-lived experience: what they've done, what it cost them, what they've already paid to fix. Facts about the past are reliable; opinions and promises about the future are not. The name comes from the idea that even your mom, who'll say yes to anything you make, couldn't mislead you if you asked the right questions.

What are the three rules of the Mom Test?

One, talk about their life, not your idea, so you never contaminate the conversation with your pitch. Two, ask about specifics in the past, not generics about the future, because people can't reliably predict their own behavior. Three, talk less and listen more, letting the customer speak the large majority of the time. Follow all three and the person can't hand you useful-sounding but worthless flattery.

Why are compliments useless in customer interviews?

Because compliments cost the other person nothing and are usually given to be kind or to end the conversation, not because they've weighed buying from you. "I'd totally use that" is free to say and predicts nothing. Chase commitments instead: a referral, a follow-up meeting, a deposit, or a signed letter of intent. Those cost time, reputation, or money, which is why they're the only reliable signal that someone will pay.

Can AI run customer interviews for me?

Use AI for the analysis, not the conversation. Have a human do the interview, then record it, transcribe it, and run it through a prompt like the one above to strip out flattery and surface the hard facts and commitments. The model is far more ruthless than you'll be at discarding the nice-sounding noise you secretly want to keep. But a live human asks the follow-ups and reads the room, and no analysis rescues an interview where you led the witness.