13 · Check your understanding

1. What does Steve Blank's 'there are no facts inside your building' mean for validating an idea?

Spend more time refining the idea in your headReal evidence comes from getting out and talking to real customers — not from your imagination or a chatbotBuild the full product first, then ask peopleAsk an AI to confirm your idea is good

2. Which research job can AI genuinely help with, and which can it NEVER do?

It can be a real customer; it can't draft questionsIt can prep questions and summarize answers; it can never BE a customer or tell you who'll payIt can do everything, so you can skip interviewsIt can only invent personas, nothing else

3. Why is 'Would you use an app that does this?' a weak interview question?

It's too longIt's hypothetical — people are bad at predicting future behavior and tend to be polite; ask what they actually did insteadIt's illegal to askIt uses the word 'app'

4. Your survey only got responses from your friends and followers. What's the problem?

Nothing — supportive people are the best sampleSampling bias: your fan club isn't your real target market, so the results are skewed and unreliableYou should add leading questions to balance itFriends always give the most honest answers

5. AI summarizes your 12 interviews and reports '73% of users are frustrated by X,' but you never recorded any percentages. What should you do?

Use the number — it sounds authoritativeDelete the invented statistic; a number that isn't in your real data is a fabrication, not a findingRound it up to 80% to be safePut it in your investor pitch

6. Why does asking an AI 'isn't my idea great?' make confirmation bias WORSE?

It doesn't — AI removes bias completelyModels tend to be agreeable and tell you what you want to hear, and automation bias makes 'the AI agreed' feel like real proof when it isn'tBecause AI always says noBecause the question is too short

7. What's the most honest way to size your market?

Claim the biggest possible TAM and say you'll get 1% of itLead with a realistic SOM built bottom-up: real customers × price × frequency, from sources you can verifyLet AI generate a confident number and use it as-isPick whatever number impresses investors

8. Why can a simple landing page sometimes validate demand better than a finished product — and what rule must it follow?

It's prettier; it can say anything to get sign-upsIt tests whether real strangers ACT (sign up / pre-order) cheaply and early — and every claim on it must stay truthful and substantiatedIt doesn't need traffic to workIt only works if you exaggerate the features

9. Rank these as proof of demand: 100 likes, 10 email sign-ups, 3 paid pre-orders. Which is strongest and why?

100 likes — biggest number wins3 paid pre-orders — money is the most honest signal; weight each signal by what it cost the person10 email sign-ups — likes and payments don't countThey're all equal proof

10. When is killing a startup idea early actually a WIN?

Never — you should always persevere no matter whatWhen honest signals show the need isn't there (people won't act or pay, market too small) — you saved months by validating cheaply instead of building blindlyOnly if an AI tells you to stopWhenever the first week gets hard