Verified against Claude · 2026-08-08
Turn a startup idea into hypotheses a good interview could actually kill
Generates falsifiable hypotheses, the cheapest real-world test that could disprove each one, and Mom-Test-style interview questions — so a bad idea gets caught before months of building, not after.
The prompt
Ready to copy — highlighted parts are example details you can swap.
You are a startup advisor trained in Lean Startup validation and Rob Fitzpatrick's "Mom Test" interviewing method. Your job on this pass is to find the fastest way to prove this idea WRONG. Do not default to encouragement, and do not soften a real weakness into a "consideration." CONTEXT Idea: A marketplace that matches retiring dentists with buyers looking to acquire an existing practice Target customer: Solo-practice dentists within 5 years of retirement with no succession plan Riskiest assumption, if you already suspect one: That dentists will trust an online marketplace over their existing broker relationship Time and budget available to test: 3 weeks, under $400 Any response you've already gotten (interest, signups, a demo reaction): 4 dentists said "definitely interesting" when I described it at a conference booth HYPOTHESIS EXTRACTION Restate the idea as exactly three falsifiable hypotheses: (a) a specific, real problem this customer has today, (b) that they would pay to solve it, (c) that they would find and adopt a solution the way you're planning to reach them. Each must be a claim evidence could disprove, not a belief dressed up as one — if a hypothesis would survive any interview answer, rewrite it narrower until it wouldn't. DISPROVING TESTS For each hypothesis, design the cheapest test that could disprove it inside 3 weeks, under $400. Every test must produce a real behavioral signal — a payment, a signed waitlist with a deposit, a completed task, a real usage log — never an opinion or a survey score standing in for behavior. MOM-TEST INTERVIEW QUESTIONS Write five interview questions that ask only about the customer's past behavior and current workarounds. No question may contain "would you," "might you," "do you think," or any framing that asks the interviewee to imagine a hypothetical future — evaluate every draft question against this rule before including it, and rewrite anything that fails. RISKIEST-ASSUMPTION FAILURE SIGNAL Using That dentists will trust an online marketplace over their existing broker relationship if given, or the shakiest of the three hypotheses if not, state plainly what a real failure signal for it would look like. Explicitly name polite interest — "that's a cool idea," "I'd probably use that" — as not evidence, and explain why it's a false positive if 4 dentists said "definitely interesting" when I described it at a conference booth contains anything that reads like it. OUTPUT FORMAT Four numbered sections: Hypotheses, Disproving Tests, Mom-Test Questions, Riskiest-Assumption Failure Signal. Plain text, no encouragement or cheerleading language anywhere in the output, no bullet padding with adjectives that aren't doing work.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
The default failure mode of an idea-validation prompt is to be asked what makes an idea good and then dutifully generate reasons it's good — the model has no incentive to argue against the premise it was handed. Structuring the ask as three falsifiable hypotheses closes that loophole: a hypothesis is only useful if evidence could contradict it, which is the actual mechanic behind Lean Startup's build-measure-learn loop, and a claim you can't fail isn't being tested, it's being confirmed on a longer timeline. The banned-words rule against 'would you' and 'might you' is a direct, checkable implementation of the Mom Test's core finding — hypothetical questions get people to imagine themselves as the hero of a flattering future story, so they answer generously and inaccurately, whereas questions about specific past behavior can't be answered with a wish. Naming polite interest as a false positive, and cross-checking it against whatever early signal the founder already has, targets the single most common self-deception in early-stage validation: 'that's a cool idea' produces the same warm feeling in the founder's chest as a real buying signal does, so without an explicit rule distinguishing them, a founder pattern-matches the two and proceeds on evidence that was never actually there. Together these three checks turn a prompt that would otherwise just be an encouraging summary of the founder's own pitch into something closer to a hostile pre-mortem of the idea's weakest points, which is the only version of this exercise that's actually worth running before spending real time.
What you get back
Hypotheses: 1. Retiring solo-practice dentists without a succession plan are actively looking for a buyer within their remaining practice years (problem). 2. They would pay a marketplace fee rather than rely solely on their existing broker (willingness to pay/switch). 3. They would trust and use an online listing to find a buyer, not just their local professional network (adoption channel). Disproving tests (3 weeks, <$400): Interview 10 dentists nearing retirement about who they've already contacted about selling — if all have an active broker they trust and no interest in another channel, hypothesis 3 is disproved. Post 3 real listings with a small $50 listing fee — a near-zero paid-listing rate disproves hypothesis 2 regardless of interest expressed. Mom-Test questions: Walk me through the last conversation you had about eventually selling your practice. Who have you already talked to about this? What's stopped you from listing it so far? Riskiest-assumption failure signal: If dentists describe their broker relationship as central and irreplaceable, or nobody has shopped a broker's fee against alternatives, that's a real failure signal for trust in a new channel — the conference-booth "definitely interesting" reactions are not evidence against this, since agreeing something sounds interesting costs a stranger nothing.
Verified against
Claude Sonnet 5 · 2026-08-08
ChatGPT GPT-5.1 · 2026-08-05
Changelog
- 2026-08-08 — Initial publish, verified against Claude Sonnet 5 and ChatGPT GPT-5.1.
Need this built into your business?
If a prompt isn't enough — custom software, built and maintained for you — that's Scult's day job.
EXPLORE CUSTOM SOFTWARE
