Bullshit Meter
Account
Field guide

How to validate a startup idea (without asking people who like you)

Startup idea validation is the practice of testing the riskiest claims underneath an idea against real evidence before investing more in building it. It is not collecting encouragement, and it is not a survey of friends: it is deliberately trying to prove your own idea wrong, cheaply, so the market does not do it expensively later.

This is the method behind our audit pipeline, written out so you can run the manual version. We built the meter in 2026 around exactly these steps, and ran it on our own product before charging anyone; it found our category story was the weak part (Bullshit Index 41, printed on the homepage). The method works on its makers.

Key takeaways

The seven steps

  1. 01Write down every claim your idea depends on. Not the pitch, the claims underneath it: who has the problem, how painful it is, why existing options fail, why you can build it, why anyone would pay. If a sentence in your pitch would collapse when one of these is false, it belongs on the list. Most founders have never seen their own claim ledger, and it is uncomfortable the first time.
  2. 02Rank claims by damage, not by ease of checking. Order the ledger by one question: if this claim is false, does the company die? Founders naturally test the claims that are easiest to confirm; the discipline is testing the one whose failure is fatal. That claim is almost never about the product. It is usually about whether the problem is painful enough for anyone to switch.
  3. 03Run counter-searches, not confirmation searches. For each fatal claim, search for the evidence that would disprove it: the competitor that already does this, the forum thread where users say the problem does not bother them, the graveyard of startups that tried. Searching to confirm what you believe is how you end up sure; searching to refute it is how you end up right.
  4. 04Interview people who owe you nothing. Friends round up because they like you. Ask strangers from your target market, and ask about their past behavior, not their opinion of your idea: what they currently do, what they last paid for, what they tried and abandoned. Rob Fitzpatrick's Mom Test rule applies: if the question can be answered with a compliment, it is the wrong question.
  5. 05Ship the smallest artifact that can prove you wrong. A landing page with a price, a concierge version, a fake-door signup: the artifact only has to be real enough to test the fatal claim. You are not building the product; you are buying evidence at the lowest available price. If the artifact cannot fail, it is not a test.
  6. 06Set kill criteria before you see results. Decide in advance what result means stop: fewer than X signups from Y visitors, fewer than X of 10 interviewees describing the problem unprompted. Criteria set after the results arrive always bend to fit them. Written-down kill criteria are the cheapest integrity device that exists.
  7. 07Re-test after every major change. A pivot invalidates old evidence. New segment, new price, new promise: the claim ledger changed, so the validation has to run again. Validation is not a gate you pass once; it is an instrument you keep re-reading while the thing is being built.

Where this goes wrong

Asking the people closest to you. Your friends, your cofounder, and the AI assistant that helped you build the thing all have the same defect: they agree with you. Enthusiasm is not evidence, and an echo is not a check.

Building more instead of testing. Adding features feels like progress and postpones the scary question. If you notice you are building to avoid finding out, stop and run step five.

Grading your own homework. After enough hours invested, you will read ambiguous evidence as encouraging. That bias is the reason kill criteria exist, and the reason we built an adversarial audit instead of another cheerleader.

Run the automated version freeNext: ten pitch red flags