Idea evaluation fails when founders fall in love with the idea and only hire AI to confirm them. These 12 prompts force honest critique — market, moat, distribution, unit economics, why-now — and end with a kill criterion in writing. Run them before you write code, before you write a deck, and before traction emotions cloud the read. Once an idea survives the gauntlet, you usually need a pricing read next — that’s where the pre-launch pricing hypothesis workflow takes over.
TL;DR
- Paste each prompt as-is and replace the
[bracketed]placeholders with your real numbers — vague inputs produce vague critiques. - Force the model into VC-partner mode, not cheerleader mode; the wording in prompt 1 does most of that work.
- Run the analytical prompts (market sizing, moat, unit economics, pre-mortem) on a strong reasoning model: Claude Opus 4.7 or GPT-5.5 Thinking as of June 2026. Claude Opus 4.7 ships a 1M-token context as standard; GPT-5.5 reaches 1M via the API (the in-app window on ChatGPT Plus is much smaller), so for pasting long interview notes and competitor pages alongside the idea, Opus 4.7 or the top ChatGPT Pro tier gives you the most room.
- Write the kill criteria (prompt 7) before you start, not after the numbers come in. That is the entire point.
Best for
- Pre-build validation
- Pre-fundraise sharpening
- Pivot decisions when traction is flat
- Side-project / weekend-build filter
- YC / accelerator application prep
Which model to run these on (June 2026)
These prompts ask for honest, multi-step reasoning, not just prose. Pick accordingly:
| Prompt type | Recommended model | Why |
|---|---|---|
| Market sizing, unit economics, pre-mortem (chains of assumptions) | Claude Opus 4.7 or GPT-5.5 Thinking | Best at flagging which assumption breaks the model; Opus 4.7 gives a full 1M-token context (GPT-5.5 hits 1M via API; smaller in-app on Plus) |
| Moat, distribution, competitor audit (needs current market facts) | Gemini 3.1 Pro or GPT-5.5 with web search | Live data on incumbents, pricing, and channels |
| Interview script, critique drafting (fast, high-volume) | GPT-5.5 Instant or Claude Sonnet 4.6 | Cheap, quick, good enough for first-draft text |
On free tiers, Claude Free (limited Sonnet 4.6) and ChatGPT Free (GPT-5.5, tight limits) both handle a single critique. For a full idea workout in one sitting, a paid tier (ChatGPT Plus $20/mo or Claude Pro $20/mo as of June 2026) removes the rate-limit interruptions. New to picking? See ChatGPT vs Claude vs Gemini.
1. One-paragraph brutal critique
My startup idea: [description]. Write a 200-word critique. Required: (a) realistic market size in one line with the assumption that drives it, (b) who actually pays and what they pay today instead, (c) why now (and why not 3 years ago / 3 years from now), (d) the single biggest risk, (e) what would kill this in 6 months. Tone: honest VC partner, not cheerleader. End with a yes / no / iterate verdict and one sentence of reasoning.2. Market sizing sketch with sensitivity
Idea: [description]. Sketch market sizing: TAM / SAM / SOM with explicit assumptions stated separately for each number. Then run sensitivity: which 2 assumptions, if wrong by 2x, would change the conclusion? Flag any number you had to invent vs source. Conclude with: is this market big enough for a venture-scale outcome, a lifestyle business, or neither?3. Moat audit
For my idea [description], identify what moats could plausibly develop: (a) network effects, (b) data advantage, (c) brand / trust, (d) switching cost, (e) economies of scale, (f) regulatory, (g) embedded workflow. Rank by likelihood specifically FOR THIS idea, not in general. Most early ideas have no moat at the start — if so, say so, and describe the path to building one within 24 months.4. Distribution channel test
For idea [description], identify 5 distribution channels (SEO, paid ads, communities, partnerships, sales, content, marketplace, virality, etc.). For each: (a) cost to get the first 100 paying users, (b) scalability past 10,000, (c) which competitors / adjacent products already win this channel, (d) my unfair advantage in this channel if any. Pick the top 2 to test in the next 30 days and what "working" looks like.5. Unit economics sketch
Sketch rough unit economics for [idea]: ARPU or revenue per user / month, COGS (hosting, API costs, fulfillment, support), gross margin %, CAC estimate by channel, payback period in months, LTV / CAC ratio. State assumptions explicitly. Flag the most uncertain number — the one that, if I'm wrong by 50%, breaks the model. Compare to a healthy SaaS benchmark for this segment.6. “Why now” sharpening
For idea [description], sharpen the "why now" thesis. Identify what specifically changed in: (a) technology / model capability, (b) regulation, (c) user behavior, (d) cost structure, (e) distribution channel. The change must be recent (<=24 months) and specific (not "AI is hot"). If you can't name a real change, the answer is "it isn't now" — say so. End with a 1-sentence why-now you could open a pitch with.7. Kill criteria with deadlines
For idea [description], define 3 kill criteria. Each: (a) a specific measurable threshold (e.g., "<5% of waitlist converts to paid pilot", "interview signal: <3 of 20 prospects say this is a top-3 problem"), (b) a deadline <=8 weeks from start, (c) what I will do if it triggers (kill, pivot to [direction], double down with more funding). The point: pre-commit so I don't move the goalposts when I'm emotional.8. Pivot menu when traction is flat
Idea [idea] has been live for [N weeks] with weak traction: [paste numbers]. Suggest 4 pivots without abandoning the core insight: (a) same product, change of audience, (b) same audience, change of pricing model, (c) same audience, change of distribution channel, (d) same audience, change of core feature. For each, name the smallest 2-week test to validate and what success looks like.9. Customer interview script for validation
For idea [description], write a 30-minute customer discovery interview script. Rules: no leading questions, no pitching the idea until minute 25. Sections: (a) current state — how do they solve this today, what do they hate, (b) recent specific stories of the problem, (c) what they have tried and abandoned, (d) what they pay for adjacent solutions, (e) brief reveal + reaction, (f) "would you pay $X" handled honestly. Include the 10 questions plus probes.10. Competitor honesty audit
For idea [description], list the top 5 alternatives a customer uses today (including "spreadsheet", "an intern", "nothing"). For each: their pricing, the job they do well, where they fail. Then name what I would actually be 10x better at — not 10% — and the customer for whom that 10x matters enough to switch. If I can't name that customer, the idea is "feature, not company".11. Founder-market fit check
Background on me as founder: [paste — experience, network, what I'm uniquely positioned to see]. Idea: [description]. Assess founder-market fit honestly: (a) why am I the right person to build this — specific reasons, not "I'm passionate", (b) what I don't know that I'd need to learn or hire, (c) who else has tried this and what they had that I don't, (d) the unfair advantage I have that a generic founder wouldn't. End with a fit score 1-10 and why.12. 12-month pre-mortem
Imagine it is 12 months from now and the startup [idea] has failed. Write the pre-mortem from that future: (a) what specifically happened — the 3 things that went wrong in order, (b) what we kept telling ourselves to ignore the signal, (c) the moment we should have killed or pivoted but didn't, (d) what the team and investor calls were like at the end. Now translate each failure mode into a leading indicator we should track from week 1.How to run a full evaluation in one sitting
- Start with prompt 1 (brutal critique) to get the shape of the bear case.
- Write kill criteria (prompt 7) immediately, before you read any more flattering output.
- Run market sizing (2) and unit economics (5) on a reasoning model; these are where AI most often invents numbers, so make it flag every guess.
- Pressure-test the thesis with why-now (6), moat (3), and the competitor audit (10).
- Close with the 12-month pre-mortem (12) and convert each failure mode into a week-1 leading indicator.
Treat AI output as a first draft of the bear case, not a verdict. The numbers it sketches are a prompt for your own research, not a substitute. For the framework these prompts are modeled on, Y Combinator’s Requests for Startups is a useful external reference for what strong founder-market fit and “why now” look like in 2026.
Common mistakes
- Evaluating only the upside, never the failure modes
- No kill criteria — every result feels “encouraging” in hindsight
- Skipping distribution entirely, assuming “we’ll figure it out post-PMF”
- Believing customer interview lip-service (“I’d totally pay for that”) over behavior
- Pivoting endlessly without finishing one validation cycle
- Accepting AI-invented TAM figures as research instead of flagging them as guesses
FAQ
Which AI model should I use to evaluate a startup idea? For the analytical prompts (market sizing, unit economics, pre-mortem), use a strong reasoning model — Claude Opus 4.7 or GPT-5.5 Thinking as of June 2026 — because they are best at naming the single assumption that breaks the model. For moat and competitor work that needs current market facts, Gemini 3.1 Pro or GPT-5.5 with web search pulls live data. A single critique runs fine on the free tier of either ChatGPT or Claude.
Can AI actually validate a startup idea? No. AI can pressure-test reasoning, surface failure modes, and draft an interview script, but it cannot tell you whether real customers will pay. It often invents market-size numbers, so treat every figure it produces as a guess to verify, not a result. Validation still requires talking to customers and watching behavior, not lip-service.
What is a kill criterion and why write it first?
A kill criterion is a specific, measurable threshold with a deadline that, if missed, means you stop or pivot — for example, <5% of the waitlist converts to a paid pilot within 8 weeks. You write it before you start because founders quietly move the goalposts once they are emotionally invested. Prompt 7 makes the model commit it to writing.
Do these prompts work on the free tier of ChatGPT or Claude? Yes for a single idea. ChatGPT Free (GPT-5.5, tight limits) and Claude Free (limited Sonnet 4.6) both handle one critique. If you want to run all 12 prompts back-to-back with long context (interview notes, competitor pages), a paid tier — ChatGPT Plus or Claude Pro, both $20/mo as of June 2026 — removes the rate-limit interruptions.
How is this different from an AI startup-validation tool? Dedicated tools (WorthBuild, IdeaProof, DimeADozen and similar) wrap a model around scraped data — Google Trends, Reddit, funding databases — and output a report. These prompts give you direct control over the critique and keep your idea private inside a model you already pay for. Use the tools for breadth of data; use these prompts for depth and honesty.