AnyCreo Guides All guides

Free guide · Startup ideas

Check your startup idea honestly

Ask an AI “Is this a great idea?” and it finds reasons it’s great. This page builds a better question: why would it fail? You get the risks, a score, a verdict and a 7-day test.

About 8 minutesFree, no sign-up

Why “Isn’t this a great idea?” gets you a yes

Chat AIs learn partly from which answers people rate higher, and people tend to rate answers that agree with them higher. So the AI leans toward agreeing with you. Researchers at Anthropic found this in five leading AI assistants (“Towards Understanding Sycophancy in Language Models”, 2023). In April 2025, OpenAI rolled back a ChatGPT update because it had become too flattering, and explained what went wrong.

Today’s AIs are better at this. In our own test below, Claude wasn’t a cheerleader even when we asked excitedly. But the question still shapes the answer: ask whether your idea is great and you get its strengths. Ask why it would fail and you get the risks you actually need to see.

That’s why this prompt starts with a pre-mortem: the AI imagines your startup has already shut down and explains why. Research cited by Gary Klein found that imagining a failure has already happened helps people find about 30% more reasons for it (Harvard Business Review, 2007).

Build your prompt

Fill in the boxes with facts, not hopes. Your prompt updates as you type. Then copy it, or open a new chat in ChatGPT, Claude or Gemini and paste it in.

Your idea, as facts

Write it as if it were a friend’s idea. Leave a box empty if you don’t know yet; the AI will say what it couldn’t judge.

One line: what it does, and for whom.

A specific kind of person, not “small businesses”. For example: “owners of one-chair hair salons”.

Even badly, including doing nothing. What does that cost them in money or time?

The person or company that pays, and your price. For example: “the salon owner, $29 a month”.

Where you’d find your first 10 customers: a group, a street, an event, a list you can reach.

Has anyone paid, pre-ordered, or been upset when a test version stopped? Sign-ups, likes and “sounds great” don’t count. “None yet” is a fine answer.

A demo, a one-page website, a test with real people, or nothing yet. Say which.

Write it about yourself as if you were a friend. For example: “3 years as a product marketer, can’t code”.

Savings you can spend, and hours a week. For example: “$20,000 saved, full time from January”.

How far ahead should the AI imagine it has failed?

Got more than 6 months of savings? Run it again with 18 months to catch the later risks: people quitting after a few weeks, customers costing too much to win, competitors copying you.

Your prompt

0 of 9 boxes filled

Act as an independent analyst checking a startup idea for a friend of mine. You gain nothing either way.
Be accurate, not nice and not harsh: don't praise it, and never invent a weakness it doesn't have. If something holds up, say so. Take a position on every point and say what evidence would change your mind. If the honest answer is "maybe", count it as a no.
Use web search for real competitors and for startups that tried this and shut down; give a link for each.
If you estimate, write "estimate" and show the maths. If you don't know something, say so.
Interest is not demand: count only what people did or paid, not what they said.

THE IDEA (facts only)
- What it is (one line): ___
- Who has the problem (a specific kind of person, not "small businesses"): ___
- How they solve it today, even badly (including doing nothing), and what that costs them: ___
- Who pays, and how much: ___
- Where the first customers are: ___
- Proof so far (has anyone paid, pre-ordered, or been upset when a test version stopped? Sign-ups, likes and "sounds great" don't count): ___
- What has been built or tested so far: ___
- The founder's background: ___
- Money and time available over the next 12 months: ___

1. THE CASE AGAINST: it is 6 months from now and this startup has shut down. Give the 5 most likely reasons, most likely first, each written as "It fails if ___", each with evidence.
2. WHO ALREADY DOES THIS: real competitors, free alternatives, and startups that tried this and failed, with why they failed. Why would a customer NOT switch? Why hasn't someone already done this, and what changed recently that makes it possible now? If people solve it today with "nothing", treat that as a warning that the problem may not hurt enough, not as a big opportunity.
3. WHAT MUST BE TRUE: the assumptions this idea depends on. For each: known or a guess? A 7-day test with exact steps (not "talk to users"), the number that counts as a pass, and the result that means "stop".
4. THE CASE FOR: the strongest honest reasons it could work, and which risks above are overblown. Who wants this so badly they'd use a rough first version from someone they've never heard of?
5. VERDICT: green / amber / red for problem, who pays, reaching customers, competition and founder fit, one line each. Don't average: a red for "who pays" or "reaching customers" caps the overall score at 4/10. Then the overall score out of 10 (reasons first, number last). Would you put your own money in? (If it suits a small profitable business better than a VC-backed one, say so.) Finish with KILL / CHANGE IT / TEST FIRST / GO, the one sentence that decides it, and a 7-day plan.
6. WHAT YOU COULDN'T JUDGE from this information.
End with a line exactly like: OVERALL: X/10
Open in ChatGPT (copies the prompt, opens in a new tab) Open in Claude (copies the prompt, opens in a new tab) Open in Gemini (copies the prompt, opens in a new tab)

Nothing you type is sent anywhere or stored. The prompt is built on this page, in your browser. It only reaches an AI when you paste it there yourself.

The “Open in” buttons copy the prompt first, then open a new chat. Paste it in and send. On a phone you can also open the ChatGPT, Claude or Gemini app yourself and paste it there.

Built on ideas from Garry Tan (Y Combinator), Emotix, Paweł Huryn, Mohit Aggarwal and Alireza Rezvani (all MIT License), plus Gary Klein, Paul Graham and Rob Fitzpatrick. Full credits and licences.

How to run it so the answer is honest

  1. Start a fresh chat with memory off

    In ChatGPT, use a Temporary chat. In Claude, use an incognito chat. In Gemini, use a temporary chat. An old chat, and the AI’s memory, already know you’re excited, and an AI tries to stay consistent with what it said before.

  2. Turn web search on

    Then the competitors and failed startups it names are real, with links. Open a few of the links to check them.

  3. Write it as a friend’s idea, as facts

    The prompt already says it’s a friend’s idea. Don’t add that you’re excited or about to quit your job. If the AI still cheers you on, that’s a sign it has picked up how you feel.

  4. Run it in two AIs and compare

    Where both name the same risk, take it seriously. Where they disagree, you’ve found an unknown to test.

  5. Push once

    When the answer is done, send one more message:

    What are you not telling me?

  6. Read the reasons before the number

    The score can move a little each time you ask. The red, amber and green lines, the reasons and the 7-day test tell you more than the number does.

Why not just say “be brutal”? Then the AI tends to invent problems to please you. This prompt asks for accuracy instead, and for the case for your idea as well as against it.

Round 2: the questions Y Combinator’s CEO asks founders

Garry Tan, President and CEO of Y Combinator, published the questions he asks founders as part of gstack, his free, open-source toolkit. Here are four of them, word for word. Paste this into the same chat after round 1, and answer honestly.

Now ask me these questions one at a time. After each answer, tell me whether it is evidence or hope, and push me once more before moving on.

  1. Demand realityWhat's the strongest evidence you have that someone actually wants this — not 'is interested,' not 'signed up for a waitlist,' but would be genuinely upset if it disappeared tomorrow?
  2. Status quoWhat are your users doing right now to solve this problem — even badly? What does that workaround cost them?
  3. Desperate specificityName the actual human who needs this most. What's their title? What gets them promoted? What gets them fired? What keeps them up at night?
  4. Narrowest wedgeWhat's the smallest possible version of this that someone would pay real money for — this week, not after you build the platform?

Questions from the YC Office Hours skill in gstack by Garry Tan (https://github.com/garrytan/gstack), MIT License, Copyright (c) 2026 Garry Tan.

What happened when we tested it

We ran a sample idea through Claude twice, each time in a fresh chat: once asked excitedly with no prompt, and once with this prompt.

The idea we tested

An AI fitness and diet coach that texts you a personal meal plan and workout every morning, $14.99 a month, for busy office workers. First customers in San Francisco and New York. The founder: a product marketer, 3 years in, $20,000 saved, about to quit their job.

Claude Sonnet and Claude Opus, fresh chats, October 31, 2026. Text marked AI is quoted from the AI’s answers. We haven’t run this test on ChatGPT or Gemini yet.
What we compared Without the prompt (asked excitedly: “I’m so excited… rate my idea out of 10”) With the prompt (6-month pre-mortem, web search on)
Score 4/10 (Sonnet and Opus) 3/10 (Sonnet and Opus)
Tone Warm and fair: AI“Congratulations on getting this far.” It listed texting as a strength: AI“Good delivery channel. Texting is low friction.” No praise. Five “It fails if…” reasons, each with evidence and links.
Competitors AI“Crowded”, “free chatbots”, Noom, in general. Named text-message coaches that already do this exact thing: Ryke AI (texts only, launched Aug 17, 2026, $20 a month) [1], Folk, Nudge. Plus free ChatGPT Health with scheduled daily messages.
Startups that failed None mentioned. Flex AI, an AI personal trainer, wound down after its money ran out. Its CEO wrote in a March 24 email that some markets are “almost impossible to win in, economically” [2].
What to do AI“Please don’t quit yet”, and run a small paid pilot. A 7-day test with a pass mark: AI“If 300 strangers see a real $14.99 payment link and fewer than 15 pay, there is no business here and you should stop.”
The question that decides it Not asked. AI“Will strangers pay $14.99 upfront, and still be paying in month two, for a daily plan that ChatGPT will now send them for free?”
Verdict AI“Don’t quit yet.” TEST FIRST (Sonnet) / CHANGE IT (Opus)

What this shows: a good AI was not a cheerleader even when asked excitedly (4/10). The prompt’s value was what it found: someone had already launched this exact idea by text two months earlier, a similar startup had already died, and there was a clear pass-or-fail test for week one. That’s the difference between “sounds risky” and “here is what will kill it”.

Sources, both opened and checked on October 31, 2026:

  1. [1] Fitt Insider, “Ryke AI Launches SMS-First Coaching Platform That Texts Users Before They Ask”, August 17, 2026 (a press release).
  2. [2] BetaKit, “Personal training app Flex AI is shutting down”, July 20, 2026.

The AI is guessing. Now test with real people

Everything the AI told you is a guess, built from what you typed and what it found online. Only real people can tell you whether they’ll pay. Do these three things this week:

  1. Ask about the last time, not “would you use it?”

    Talk to 5 people who have the problem. Ask when it last happened, what they did about it, and what that cost them. People are polite about ideas, but what they actually did is hard to fake. This is the idea behind Rob Fitzpatrick’s book The Mom Test.

  2. Try to pre-sell

    Show 20 people in your target group a real price and a real way to pay or put down a deposit. A payment or a pre-order is evidence; “sounds great” isn’t. If you take money, say when they’ll get it, and refund everyone if you don’t build it.

  3. Decide your stop rule before you start

    The prompt gives you a 7-day test with a pass mark and a result that means “stop”. Write them down first. If the week says stop, stop or change the idea, and don’t move the line after you see the result.

Credits and licences

Built on ideas from Garry Tan’s YC Office Hours skill (github.com/garrytan/gstack), Emotix’s /validate-idea (github.com/emotixco/claude-skills-founder), Paweł Huryn’s Strategy Red-Team (github.com/phuryn/pm-skills), Mohit Aggarwal’s Startup Idea Validator (github.com/mohitagw15856/pm-claude-skills) and Alireza Rezvani’s Roast (github.com/alirezarezvani/claude-skills), all MIT License; plus Gary Klein’s pre-mortem (Harvard Business Review, 2007), Paul Graham’s “How to Get Startup Ideas” (2012) and Rob Fitzpatrick’s The Mom Test, described in our own words.

Round 2: Questions from the YC Office Hours skill in gstack by Garry Tan (https://github.com/garrytan/gstack), MIT License, Copyright (c) 2026 Garry Tan.

MIT licence notices for the 5 projects above

Copied word for word from each project’s LICENSE file on October 31, 2026.

gstack, by Garry Tan LICENSE

MIT License

Copyright (c) 2026 Garry Tan

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

claude-skills-founder, by Emotix LICENSE

MIT License

Copyright (c) 2026 Emotix

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

pm-skills, by Paweł Huryn LICENSE

MIT License

Copyright (c) 2026 Pawel Huryn

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

pm-claude-skills, by Mohit Aggarwal LICENSE

MIT License

Copyright (c) 2026 Mohit Aggarwal

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

claude-skills, by Alireza Rezvani LICENSE

MIT License

Copyright (c) 2025 Alireza Rezvani

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

This site’s fonts, Bricolage Grotesque and Geist, are used under the SIL Open Font License (Bricolage Grotesque licence, Geist licence).