The Mom Test in the AI Era: What Still Works in 2026

The ideas here draw on Customer-Led Development: How to Build What Your Customers Want When AI Can Build Anything — the operating manual for building what customers want when AI can build anything. See every chapter.

You can now build the thing faster than you can schedule the interview about the thing. That one shift quietly breaks half the advice on customer discovery and leaves the other half more important than ever.

Rob Fitzpatrick's The Mom Test (2013) is still the best book ever written about a single customer conversation. Nothing in it has expired. But it was written for a world where building was expensive and talking was cheap, so the conversation was the main event. In 2026, the conversation is one cheap test among several. The bottleneck has moved from asking to deciding and following through.

This is a guide to what carries over, what to add, and the new way teams fail even when they interview well.

The Mom Test in one minute

The premise: everyone, especially people who like you, will lie to you about your idea. Not maliciously. They're being kind. Your mom will say it sounds great. So your job is to ask questions that even your mom couldn't lie to you about.

Fitzpatrick boils it down to three rules (Paraphrased.):

Talk about their life, not your idea. The moment you pitch, you get compliments instead of information.
Ask about specifics in the past, not generalities or the future. "When did this last happen?" beats "Would you use this?"
Talk less, listen more. If you're talking, you're not learning.

Around those rules sit a few supporting ideas worth memorizing (Paraphrased.):

Compliments are worthless. They feel like data and contain none.
Watch for fluff: generic claims ("I usually..."), future promises ("I would definitely..."), and hypotheticals ("I might..."). Pull them back to concrete past behavior.
Feature requests are clues, not instructions. Dig for the motivation behind them.
Look for commitment and advancement. A good meeting ends with the other person giving up something real: time, reputation, or money.
Keep it casual. A coffee chat yields more truth than a formal "interview."

If you only ever do those things, you'll be better at customer conversations than most product teams on Earth.

What still works exactly as written

Some advice ages. This doesn't, because it's about human nature, and human nature didn't get a model update.

People still lie to be nice

AI didn't make customers more honest. If anything, a slick prototype makes the problem worse: it's easier to compliment something that looks finished. "Wow, this is cool" is the 2026 version of "sounds great, honey."

Past behavior is still the only reliable signal in a conversation

"Tell me about the last time you had to reconcile those invoices" still beats every hypothetical. What someone did last Tuesday, what they paid for, what workaround they built in a spreadsheet: that's the evidence. Opinions about the future are still noise.

Commitment is still the real test

A meeting that ends with "keep me posted" was a nice chat. A meeting that ends with an intro to their boss, a scheduled pilot, or a deposit was a data point. This is the part of the book most people skip, and it's the part that matters most once building gets cheap.

Rule of thumb: if the conversation cost them nothing, assume you learned nothing about demand.

What AI changes

When the book came out, building a working product to test an idea took months. So you squeezed everything you could out of conversations first. That math is gone. A competent person with AI tools can put a working prototype in front of users over a weekend.

That changes three things.

1. The interview is no longer the only cheap test

The old sequence was: talk, talk, talk, then build once. Now you can interleave. Talk to five people on Monday, ship something rough on Wednesday, watch what they do with it on Friday. The interview isn't replaced; it's joined by tests that measure behavior directly.

This is good news for Mom Test purists. The book's whole point is that behavior beats opinion. A prototype someone actually uses (or ignores) is behavior.

2. Prototypes become questions

The trap is treating a prototype as an answer: "Here's the thing, do you like it?" That's just pitching with extra steps, and it invites the exact compliments Fitzpatrick warned about.

Instead, build each prototype to answer one specific question:

Will they connect their real data, or just poke at the demo?
Do they come back on day three without a reminder?
When it breaks, do they complain (they care) or vanish (they don't)?
Will they pay, pre-order, or put their name on a waitlist that requires effort?

Then run the conversation around the prototype using Mom Test rules. Don't ask what they think of it. Ask what they were trying to do when they opened it, and what they did instead last time.

Rule of thumb: a prototype without a written question is a demo, and demos collect compliments.

3. Production becomes research

Once something is live, every support ticket, sales call, cancellation reason, and usage log is a customer conversation you didn't have to schedule. Teams that once ran twelve interviews a quarter now sit on thousands of hours of raw customer signal.

That sounds like a solved problem. It isn't. It's a new one.

The new failure: talking to customers and still not acting

The classic failure The Mom Test fixes is asking badly: leading questions, pitching, collecting fluff. The 2026 failure is subtler. Teams ask well, hear clearly, and then nothing happens.

The numbers on this are uncomfortable. In a McKinsey survey of more than 500 senior executives, 63% named customer feedback as a top source of growth ideas, but only 15% said they consistently use customer input in their decisions. That gap isn't an interviewing problem. No amount of better questions closes it.

The cost of the gap shows up in what gets built. Pendo's 2019 Feature Adoption Report put roughly 80% of features in the rarely-or-never-used bucket, with waste estimated at up to $8.4 million a year for a $50M company. AI doesn't fix that. AI makes it cheaper to ship the wrong thing faster.

The Intuit story, both halves

Intuit is the best example of getting the conversation right and the follow-through wrong, then right again.

In its early years, Intuit employees would wait in computer stores until someone bought Quicken, then ask to follow the buyer home and watch the install. They noted everything, from the shrink-wrap to the instructions that produced confused looks, and co-founder Scott Cook treated every stumble as Intuit's fault (Inc.). That's the Mom Test in its purest form: observe real behavior, in context, with zero pitching.

Then the second half. Intuit's own surveys showed many Quicken customers were using the personal finance program at work to keep their company's books. The signal was there, in the data, from real customers. Cook said Intuit ignored it for four years. When the company finally built QuickBooks for those customers, the business became, in Cook's words in 2014, 11 times larger than Quicken (Inc.).

Same company, same customer obsession, same quality of listening. The difference between the two halves wasn't asking. It was acting.

The Mom Test teaches you to get the truth out of a conversation. It doesn't tell you what to do once you have it, and in 2026, that's where most of the value leaks out.

From one good conversation to a system

Fitzpatrick's book is about the conversation. Spencer Shulem's Customer-Led Development picks up where the conversation ends: how customer truth turns into decisions, and how those decisions get traced back to the people who drove them. Its core distinction is useful here.

A customer-influenced company has customer input somewhere in the process: surveys, a feedback portal, PMs who talk to users sometimes, NPS on a dashboard. A customer-led company has customers driving the process: features have names attached before development starts, customers are told when their request ships, and feedback-to-decision time is measured in hours, not months.

Most teams who've read The Mom Test are customer-influenced. They have great conversations. The conversations just don't connect to anything.

Three practices close that gap.

1. Named customers, not personas

The test from the book is blunt: can you name the specific customers behind your last strategic decision? Not "mid-market fintech." Not "our ICP." Names.

This is the Mom Test's specificity rule applied one level up. Fitzpatrick tells you to stop asking about generic behavior and ask about the last specific time. The same discipline applies to decisions. "Users want better reporting" is fluff. "Acme, TechCo, and BigBrand each told us they rebuild this report by hand every month" is evidence.

Rule of thumb: if you can't name the customers behind a feature, you're building on intuition with extra steps.

2. Promises you write down

Every good customer conversation ends with an implicit promise: "Thanks, this is really helpful, we'll look into it." Most teams never track those promises. The customer remembers. You don't.

Treat each promise as a commitment in the Mom Test sense, running the other direction. They gave you time and candor; you owe them a response. Log who said what, what you told them, and what you'll do.

3. Closing loops

When the feature ships, the customers who asked for it should hear about it from you, by name, not through a generic release note. A call, a short personal video, a direct email. "You told us X in March. Here's what we built. Does it actually solve it?"

That last question is the point. Closing the loop isn't customer service; it's the validation step. It turns every shipped feature into a follow-up interview with someone who already has context and a reason to be honest.

A quick self-test

Adapted from Customer-Led Development. Pick your last five major decisions and ask, for each:

Can you name the specific customers who drove it?
Can you find the exact quotes or feedback that led to it?
Is there a plan to close the loop with those customers when it's done?

Yes to all three for at least four of five, and you're customer-led. Fewer, and you're somewhere on the spectrum between influenced and led. That's a diagnosis, not a judgment.

The Mom Test rules, updated for 2026

Mom Test rule (paraphrased)	Still true?	What to add in 2026
Talk about their life, not your idea	Yes, fully	Applies to prototypes too. Don't ask what they think of it; ask what they were trying to do when they used it.
Ask about specific past behavior, not the future	Yes, fully	Pair interviews with real usage data. Past behavior now includes what they did in your product yesterday.
Talk less, listen more	Yes	Also: read more. Support tickets, sales calls, and churn reasons are conversations you already had.
Compliments are worthless	More true than ever	Polished AI prototypes attract more compliments. Discount reactions to demos heavily.
Feature requests are clues, not instructions	Yes	Log the request with the customer's name attached, then dig for the motivation. You'll need both later.
Push for commitment and advancement	Yes, and now testable	A working prototype lets you ask for real commitment (data, payment, a pilot) far earlier.
Keep it casual	Yes	Casual conversations, rigorous records. Informal tone, formal follow-through.
(Not in the book) Act on what you hear	N/A	Attach names to decisions, track promises, close loops when you ship. This is where most value now leaks.

A 2026 discovery loop in practice

Here's what a week looks like when you combine Mom Test conversations with cheap building and real follow-through:

Monday: five Mom Test conversations. Past behavior only. No pitching. Write down names and exact quotes.
Tuesday: pick one question. Not "is this a good idea?" but "will they connect their real accounts?" or "will they come back without a nudge?"
Wednesday: build the smallest thing that answers it. With AI tools, this is often a day, not a sprint.
Thursday–Friday: watch, then talk. Look at what they did before you ask what they think. Ask for a real commitment.
Every week: decide with names. Each build-or-kill call cites specific customers and their quotes.
On ship: close the loop. Tell the named customers, ask if it solved the problem, log the answer.

Notice what the interview is doing here. It's no longer a gate you pass through once before building. It's the connective tissue between every cheap experiment.

FAQ

What is The Mom Test in one sentence?

It's Rob Fitzpatrick's method for asking customers questions about their real, past behavior so that even someone who wants to be nice to you can't give you a misleading answer.

Is The Mom Test still relevant with AI?

Yes. The rules are about human psychology, which hasn't changed. What's changed is that interviews now sit alongside cheap prototypes and live usage data, and the biggest risk has shifted from asking badly to failing to act on what you hear.

Should I build a prototype before doing customer interviews?

Talk first, briefly, to find a real problem worth testing. Then build a small prototype designed to answer one specific question, and use Mom Test-style conversations around it. Don't use the prototype as a pitch.

How many customer interviews do I need?

Fitzpatrick's guidance is to keep going until you stop hearing new things (Paraphrased.). In practice, the better question is whether you can name the specific customers behind each decision you make. If you can't, you need more conversations or better records of the ones you had.

What's the difference between customer-influenced and customer-led?

In Spencer Shulem's Customer-Led Development, customer-influenced teams have customer input somewhere in the process; customer-led teams attach customer names to decisions before building, notify those customers when things ship, and move from feedback to decision in hours rather than months.

What should I do after a good customer interview?

Log the customer's name, their exact words, and anything you promised. When a decision touches that problem, cite them. When you ship, tell them directly and ask whether it actually solved the problem.

If you want the full system for turning great conversations into named decisions and closed loops, it's all in Customer-Led Development.

And if ideas like "the hard part isn't asking anymore, it's acting" are your kind of thing, subscribe to the Context Limit newsletter for more that fit in your head.

The Mom Test in the AI Era: What Still Works and What Doesn't (2026)