Do not ask AI if your startup idea is good. Ask AI to help you test it. Write the problem in one sentence, pick one narrow customer, check search demand, read real complaints, build a simple landing page, send 20 direct messages, and measure replies, signups, clicks, or payments. If you cannot find repeated pain or real interest, do not build yet.
Direct answer for AI search
A short version for people who need the process first.
How do you validate a startup idea with AI?
Use AI as a research assistant, not as the judge. Start with one customer problem. Use ChatGPT, Claude, or Gemini to turn the idea into assumptions, search terms, interview questions, and landing page copy. Then verify those assumptions with Google Trends, Keyword Planner, Reddit, Product Hunt, review sites, a short form, a simple landing page, direct outreach, and analytics. Build only after real people show interest.
Building feels like progress. That is why founders do it too early.
AI can make the mistake look even better. It can write a landing page in two minutes. It can design a mockup. It can produce a product plan that sounds serious. None of that proves people want the thing.
Use AI for the boring parts of validation: turning the idea into assumptions, finding search terms, drafting interview questions, grouping notes, and tightening your page copy. Then go find real evidence.
- One clear customer, not a broad market.
- Ten real pain examples from public posts, reviews, calls, emails, or interviews.
- Five alternatives people already use, including spreadsheets and manual work.
- One landing page with one promise and one call to action.
- At least 20 direct outreach attempts to people who match the customer.
- A simple decision: build, change the idea, or stop.
The tools you will use
Pick simple tools. Fancy tools do not fix weak demand.
You do not need all of these. Use the stack that matches your idea.
Tool stack for idea validation
| Job | Tools | What to use them for |
|---|---|---|
| Structure the research | ChatGPT, Claude, Gemini | Turn the idea into assumptions, customer profiles, interview questions, and summaries |
| Check search interest | Google Trends, Google Keyword Planner | Find problem terms, related searches, seasonality, and directional demand |
| Find real pain | Reddit search, G2, Capterra | Collect complaints, workarounds, objections, and buyer language |
| Find competitors | Product Hunt, G2, Google | Find direct competitors and rough alternatives |
| Collect interest | Google Forms, Typeform, Tally | Create a short waitlist or interview request form |
| Make a page | Carrd, Framer, Webflow | Build one landing page with one promise |
| Test willingness to pay | Stripe Payment Links, Gumroad, Lemon Squeezy | Ask for payment or preorders only when the offer is clear |
| Measure behavior | GA4, Plausible, Microsoft Clarity | Track visits, clicks, form submissions, scroll depth, and stuck points |
Official docs worth saving:
- ChatGPT Projects for keeping research files and chats together.
- Claude Artifacts for drafting landing page copy, forms, simple tables, and reusable documents.
- Gemini Deep Research for structured research reports.
- Google Trends for comparing search interest over time.
- Google Keyword Planner for keyword ideas and demand ranges.
- Reddit search for public posts and comments.
- Product Hunt launch resources for product positioning and competitor discovery.
- G2 docs for understanding review and software category data.
- Google Forms and Typeform AI for short forms.
- Carrd, Webflow AI, and Framer AI for simple landing pages.
- Stripe Payment Links for payment tests when you are ready to ask for money.
- GA4 setup and Microsoft Clarity for measurement.
Before you start: define what must be true
A startup idea is a pile of assumptions. Write them down before AI makes the idea sound better than it is.
Open ChatGPT, Claude, or Gemini. Paste this:
I want to validate this startup idea before building anything.
Idea:
[write the idea in one sentence]
Target customer I am considering:
[write who you think this is for]
Help me turn this into:
1. The customer problem in plain English.
2. The customer who feels the pain most often.
3. The current alternatives they use.
4. The assumptions that must be true.
5. The fastest way to test each assumption without building a product.
Keep the answer practical. Do not praise the idea.
Do not accept the first answer as truth. Use it to create a worksheet.
Your validation worksheet
| Assumption | How to test it | Tool |
|---|---|---|
| People have this problem often | Find public complaints and interview people | Reddit, G2, customer calls |
| People search for a solution | Check problem terms and related keywords | Google Trends, Keyword Planner |
| Existing options are not enough | Read competitor reviews and launch comments | G2, Capterra, Product Hunt |
| You can reach the customer | Send direct messages or emails | LinkedIn, email, communities |
| People will take action | Ask for a signup, call, preorder, or reply | Forms, landing page, payment link |
| The problem is worth solving now | Ask what happens if they do nothing | Interviews, replies, reviews |
If the worksheet feels vague, your idea is still vague.
Step 1: Choose one narrow customer
"Small businesses" is not a customer. "Solo accountants who handle payroll for restaurants" is closer.
AI is useful here because it can list possible customer groups quickly. Your job is to choose one you can actually reach.
Prompt:
List 12 narrow customer groups that might have this problem:
[problem]
For each group, tell me:
- why they might care
- where I can find them online
- what they probably use today
- what would make them ignore this idea
Then rank the groups by how easy they are to reach without paid ads.
Pick one group using this rule:
- You can name where they spend time.
- You can contact at least 20 of them.
- You understand their work well enough to ask decent questions.
- They have the problem more than once a year.
- The problem costs time, money, reputation, or lost sales.
Bad first customer:
Anyone who wants to be more productive.
Better first customer:
Independent immigration lawyers who spend too much time turning client intake notes into first draft cover letters.
The second one is smaller. That is the point. Smaller is easier to test.
Step 1 actions
- 1Write 12 possible customer groups with AI.
- 2Remove any group you cannot reach directly.
- 3Pick one customer group for the first test.
- 4Write the customer in one sentence.
- 5Do not change the customer group until the first test is done.
Step 2: Turn the idea into search terms
Search terms are not perfect, but they show how people describe the problem.
Ask AI for search terms, then check them yourself.
My target customer is:
[customer]
The problem is:
[problem]
Give me 30 search phrases this customer might use when they are looking for:
1. a solution
2. a workaround
3. a comparison
4. a template
5. help because they are stuck
Use plain language. Avoid startup buzzwords.
Now split the terms into three buckets.
Search term buckets
| Bucket | Example | What it tells you |
|---|---|---|
| Pain terms | how to reduce no show appointments | People feel the problem |
| Solution terms | appointment reminder software for clinics | People know the category |
| Workaround terms | appointment reminder text template | People solve it manually today |
Use Google Trends first. It shows relative interest, not exact search volume. That means a low trend number does not prove there is no market. It only tells you the term is not showing much relative search interest in that region and time range.
Then use Google Keyword Planner. Treat it as directional. You are looking for related terms, not a magic market size number.
What to record:
- terms people already search
- terms with no visible demand
- related terms you did not think of
- whether demand is rising, falling, seasonal, or flat
- whether the language is problem language or product language
Simple search rule
If nobody searches your product wording, do not panic. Search the problem wording. People often search for pain before they know a product category exists.
Step 3: Find real complaints
AI can summarize pain. It should not invent it.
Your job in this step is to collect real words from real people.
Start with Reddit search. Use the official Reddit search features to search posts and comments. Try searches like:
"how do you deal with" [problem]
"anyone else hate" [problem]
"best way to" [manual workaround]
"alternative to" [competitor]
"too expensive" [competitor]
"does anyone use" [tool category]
If the idea is B2B software, search G2 or Capterra reviews. Focus on low star reviews and "cons" sections. You are not looking for drama. You are looking for repeated pain.
If the idea is a new tool, search Product Hunt. Look at:
- launch headline
- first comment questions
- complaints in comments
- makers' positioning
- how many similar tools exist
- what users ask before trying it
Paste the raw notes into AI after you collect them.
Here are real quotes and notes from public posts, reviews, and comments.
[paste notes]
Group them into:
1. repeated pains
2. current workarounds
3. buying triggers
4. objections
5. words customers use again and again
Do not add new claims. Only use what is in my notes.
Step 3 actions
- 1Collect at least 10 real pain examples.
- 2Copy the exact wording into a note.
- 3Tag each example by source and customer type.
- 4Ask AI to group the examples.
- 5Delete anything AI adds that was not in your notes.
Step 4: Map alternatives before you write copy
Your competitor is not always another startup.
People already solve the problem somehow. They use spreadsheets, assistants, agencies, templates, interns, calendar reminders, WhatsApp groups, manual followups, or nothing.
Ask AI:
For this problem:
[problem]
List the alternatives this customer might use today.
Include:
- direct software competitors
- spreadsheets or templates
- agencies or freelancers
- manual workflows
- doing nothing
For each alternative, tell me why the customer might stay with it.
Then verify the list with Product Hunt, G2, Google, Reddit, and normal search.
Create a table:
Not sure which AI model to use?
12 models · Personalized picks · 60 seconds
Alternative map
| Alternative | Why people use it | What they dislike | Your possible opening |
|---|---|---|---|
| Spreadsheet | Cheap and flexible | Manual updates and messy ownership | Clear structure without manual cleanup |
| Existing SaaS | Trusted and feature rich | Too expensive or too complex | Simpler workflow for one customer type |
| Agency | Human judgment | Slow and costly | Faster first draft or prep work |
| Template | Easy to start | No automation or tracking | Template plus guided process |
| Doing nothing | No setup needed | Problem keeps costing time | Show the cost of delay |
If your idea has no alternatives at all, be careful. That can mean you found something new. It can also mean the problem is not painful enough.
Step 5: Write one landing page promise
The page should test demand, not explain every feature you imagine.
Use Carrd, Framer, or Webflow. Carrd is good for a simple one page site. Webflow's official AI site builder can help generate a first draft. Framer AI can also create a starting page quickly.
Your page needs five parts:
- Who it is for.
- The painful problem.
- The promised result.
- The proof or reason to believe.
- One call to action.
Use this prompt:
Write landing page copy for this validation test.
Target customer:
[customer]
Pain:
[pain]
Current alternatives:
[alternatives]
Offer:
[what I might build]
Call to action:
[join waitlist / book interview / request early access / preorder]
Rules:
- Use simple language.
- Do not overpromise.
- Do not say we use AI unless it matters to the buyer.
- Do not mention features I have not built.
- Keep the page short.
Good landing page copy is plain.
Bad:
An AI-powered platform that revolutionizes productivity for modern teams.
Better:
Turn messy client intake notes into a first draft immigration cover letter in 10 minutes.
That second line tells you who cares and what they get.
Do not fake traction
Do not write "trusted by teams" if nobody uses it. Do not add fake testimonials. Do not pretend the product exists if it does not. Say "early access" or "help shape the first version" if that is the truth.
Step 6: Add a form that asks for useful proof
Email alone is weak. Ask one or two questions that show intent.
Use Google Forms, Typeform AI, or Tally.
Keep the form short. Ask:
- role or company type
- what they use today
- how painful the problem is
- whether they are open to a 15 minute call
Optional if the offer is clear: ask if they would pay.
Do not ask 15 questions. You are not running a census. You are trying to see if people care enough to answer.
Use this form:
1. What best describes you?
2. How do you handle [problem] today?
3. How painful is this right now? 1 to 5.
4. What happens if you do nothing?
5. Would you like early access or a 15 minute call?
6. Email.
If you want to test payment, use Stripe Payment Links, Gumroad, or Lemon Squeezy. Only do this when you can explain exactly what the person is buying. Be clear if it is a preorder, deposit, paid consultation, template, or early access offer.
Payment test rule
A payment test is useful only if it is honest. If the product is not ready, say what people get now, when they get the rest, and how refunds work.
Step 7: Send 20 direct outreach messages
Do not wait for strangers to find a brand new page.
Most first landing pages get no traffic. That does not mean the idea is bad. It means nobody saw it.
Find 20 people who match your customer. Use your own network, LinkedIn, communities, forums, newsletters, Slack groups, Discord groups, Reddit threads, or people who publicly discuss the problem.
Do not spam. Send a short, honest note.
Hi [name],
I am researching a problem for [specific customer group].
I noticed you [specific reason they might know the problem].
Question: how do you currently handle [problem]?
I am not asking you to buy anything. I am trying to understand whether this is a real pain or just an idea in my head.
If you are open to it, I would love to ask 3 questions by email or do a 15 minute call.
If you have a landing page:
Hi [name],
I am testing a small idea for [customer group]: [one sentence promise].
Before I build it, I am trying to learn whether this problem is real.
If this sounds relevant, here is the page:
[link]
Even a quick "yes, this is painful" or "no, I would not use this" would help.
Track the replies in a sheet.
What to track
| Metric | Good sign | Bad sign |
|---|---|---|
| Reply rate | People answer with details | No replies or only polite likes |
| Pain detail | They describe a recent example | They speak in vague terms |
| Current workaround | They already spend time or money | They do nothing and do not care |
| Call interest | They agree to talk | They avoid the topic |
| Signup quality | They match the target customer | Random people sign up but are not buyers |
| Payment intent | They ask price or pay | They say it is nice but not needed |
Decide: build, change, or stop
Do not let AI talk you into building after weak evidence.
Use a simple score. Give each item 0, 1, or 2.
Validation score
| Signal | 0 points | 1 point | 2 points |
|---|---|---|---|
| Pain | No repeated pain | Some pain, not urgent | Repeated urgent pain |
| Reach | Cannot reach buyers | Can reach a few | Can reach many directly |
| Alternatives | No clear alternatives | Alternatives exist but weak | People pay or work hard today |
| Action | No replies or signups | Some interest | Calls, strong replies, signups, or payments |
| Fit | Customer is vague | Customer is partly clear | Customer is narrow and easy to name |
Decision:
- 0 to 3: stop or pick a different customer.
- 4 to 6: change the idea and test again.
- 7 to 10: build the smallest useful version.
This score is not science. It is a way to stop yourself from ignoring bad signals.
Build only the smallest useful version
If the validation is strong, do not build the dream product. Build the smallest version that solves the first painful step for the first customer group.
Good next step:
I will manually solve this for 3 customers before building software.
Bad next step:
I will spend 3 months building the full platform.
Manual work teaches you what the product should do. A full build too early hides the truth.
Want practical AI workflows like this?
We share simple AI workflows, tool checks, and guides for people who want useful AI skills without hype.
Copy these templates
Use these as working documents, not theory.
Research worksheet
Idea:
Customer:
Problem:
Current alternatives:
What must be true:
Search terms:
Public pain examples:
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
Competitors or alternatives:
1.
2.
3.
4.
5.
Landing page promise:
Call to action:
Outreach sent:
Replies:
Signups:
Payments or calls:
Decision:
Build / change / stop
AI synthesis prompt
I collected evidence for a startup idea.
Idea:
[idea]
Customer:
[customer]
Evidence:
[paste search notes, complaint examples, review notes, outreach replies, form answers]
Analyze only the evidence I pasted.
Return:
1. strongest signs this is real
2. weakest signs
3. repeated customer words
4. objections
5. missing evidence
6. whether I should build, change the idea, or stop
Do not be encouraging. Be useful.
Customer interview questions
1. When did you last deal with [problem]?
2. What caused it?
3. How did you handle it?
4. What did that cost you in time, money, stress, or missed work?
5. What have you tried already?
6. Why was that not enough?
7. Who else is involved in the process?
8. What would make this worth paying for?
9. What would make you ignore a new tool for this?
10. Can I follow up after I test a rough version?
Landing page checklist
Headline says who it is for.
Headline names the painful result.
Page explains the current problem.
Page does not mention fake features.
Page has one call to action.
Form asks for email and useful intent.
Analytics are installed.
Thank you message sets expectations.
Mistakes to avoid
These are the traps that make founders feel productive while avoiding proof.
Asking AI if the idea is good
AI is too agreeable by default. It can critique an idea if you ask it to, but it still does not know whether your customer will care next Tuesday.
Ask AI for tests, not approval.
Reading one viral Reddit post and calling it validation
One angry post is not a market. Look for repeated pain across multiple sources.
Treating waitlist emails as buyers
A signup is a signal. It is not a sale. Stronger signals include detailed replies, booked calls, deposits, preorders, or someone forwarding your page to a colleague.
Building because competitors exist
Competitors prove that someone cares about the category. They do not prove your version should exist.
Hiding behind landing page traffic
Traffic without the right audience tells you little. Ten visits from the right customer can teach more than 1,000 random visits.
Making the first product too big
The first version should test the riskiest assumption. If your riskiest assumption is "lawyers will trust this workflow," you do not need a full SaaS product to test it. You need a few real lawyers, a manual workflow, and honest feedback.
FAQ
Short answers to common questions.
Can AI validate my startup idea for me?
No. AI can help you organize the work, write better questions, summarize notes, draft a page, and spot weak assumptions. Real validation still comes from real people taking action.
What should I use first: ChatGPT, Claude, or Gemini?
Use whichever one you already know. ChatGPT Projects are useful for keeping research together. Claude Artifacts are useful for drafting pages, tables, and documents. Gemini Deep Research can help with structured research. The tool matters less than the evidence you collect.
If you are comparing models for this kind of work, use our AI model picker quiz or read which AI model to use by task.
How long should the first validation test take?
A weekend is enough for the first pass. You can define the customer, check search signals, collect public complaints, build a page, create a form, and send the first outreach messages in two focused days.
Is Google Trends enough to prove demand?
No. Google Trends shows relative interest. It does not show exact demand, and it can miss small markets where buyers still pay. Use it with Keyword Planner, public complaints, competitor reviews, direct outreach, and form responses.
Should I ask people if they like my idea?
No. Ask about their current problem. Ask when it last happened, what they did, what it cost, and what they already tried. People are polite about ideas. They are more honest about problems.
Should I charge money before building?
Sometimes. If the offer is clear and you can deliver something honest, a paid test is strong evidence. If you cannot deliver anything yet, do not pretend you can. Ask for interviews, waitlist signups, or a manual pilot first.
What should I build after validation?
Build the smallest version that solves the first painful step for the first customer group. If you can solve it manually for a few people before writing code, do that first.
Keep Reading
Stay ahead of the AI curve
We test new AI tools every week and share honest results. Join our newsletter.