How to validate a startup idea in a week: 9 checks

Many startups don't fail because the founder can't build. They fail because not enough people needed the thing. CB Insights studied 431 venture-backed startups that shut down since 2023, and poor product-market fit was a cause for 43% of the ones whose reasons it could identify.

Paul Graham, the co-founder of Y Combinator, calls solving problems no one has "by far the most common mistake startups make." The good news is that you can test for that mistake before you write a line of code. AI agents can now take on much of the research, the interviews, and the landing page, which makes a one-week validation realistic for one founder.

Below are nine checks, one plan for the week, and the tools that do most of the work for each one.

What does it mean to validate a startup idea?

Validating a startup idea means collecting evidence that real people have the problem, that they want your way of solving it, and that they will pay for it. Opinions don't count. Behavior counts: a search people already type, a workaround they already use, an email they leave, or money they put down.

Treat your idea as a guess you can prove wrong. A good test has a clear question, a cheap way to answer it, and a result you decided in advance would mean "stop."

Why should you validate an idea before you build it?

Testing an idea before you build it helps you spot weak ideas and move on to better ones. A randomized controlled trial with 116 Italian startups, published in Management Science, taught one group to test their ideas the way scientists test hypotheses. Those founders performed better, were more likely to pivot to a different idea, and were no more likely to drop out than the control group.

A larger replication with 759 firms across four randomized trials, published in the Strategic Management Journal in 2024, found the same approach led more founders to end ideas that weren't working. The authors describe it as making founders more efficient in their search for a viable idea. Ending a weak idea in week one is a win. It frees you for the one that works.

What are the 9 checks to validate a startup idea in a week?

The nine checks below fit into seven days. Run them in order, because each one tells you whether the next is worth doing.

1. Day 1: Write the idea as one sentence. Name who has the problem, what the problem is, and how they solve it today. "Independent dental clinics miss calls after hours and pay a front-desk temp to cover" is testable. "AI for healthcare" is not.
2. Day 1: List your riskiest assumptions and set kill criteria. Write down what must be true for the idea to work, then decide the result that would make you stop. For example: fewer than 3 of 10 interviews describe the problem without prompting, or nobody pre-orders.
3. Day 2: Check that people search for the problem. Google Trends shows how interest in a search term changes over time and by place, scaled from 0 to 100. A rising or steady line for the problem, not your product name, is a good sign. Google Keyword Planner adds rough monthly numbers.
4. Day 2: Map who already solves it. Ask a research assistant like Perplexity, which answers with cited sources, to list the products, agencies, and spreadsheets people use today. Competitors are often good news. They suggest a budget exists, and Paul Graham writes that a crowded market is actually a good sign.
5. Day 3: Find ten people who have the problem. Look where they already talk: niche forums, subreddits, industry groups, and your own network. Pick people who match your one-sentence customer, not friends who want to be kind.
6. Days 3 and 4: Interview them about their past. Ask what happened the last time they hit the problem, what it cost them, and what they tried. Don't pitch. You learn the most when you talk the least.
7. Day 4: Widen the sample with AI interviews. An AI interviewer like Listen Labs or Outset holds the conversation, asks follow-up questions, and sums up the themes. You test your questions with real people, then let the agent run many more while you work on the next check.
8. Day 5: Put up a one-page offer. Describe the result, not the features, and add one clear action. Framer designs and publishes landing pages with AI, and Lovable builds a web page or a simple app from a chat prompt. Send the page to the people you interviewed and to the places you found them.
9. Days 6 and 7: Ask for money, then decide. A Stripe payment link lets you take a pre-order or a deposit without writing code. On day 7, score everything against the kill criteria you wrote on day 1. Build, change one part of the idea, or stop.

How do you talk to customers without getting false answers?

Customer interviews give false answers when you ask people to judge your idea. Polite people say "I'd use that" to be kind, and kindness isn't evidence. Rob Fitzpatrick's book The Mom Test is a short handbook on this problem: how to talk to customers and learn if your business is a good idea when everyone is lying to you.

A few habits keep the answers honest:

• Ask about their life, not your idea. "How do you handle this today?" beats "Would you use an app that does this?"
• Ask about specific past events. "Tell me about the last time" gets facts. "Would you ever" gets wishes.
• Ask what they've already spent. Treat time, money, or a hire spent on the problem as the strongest signal a conversation can give you.
• Listen for the pain in their words. Write down the exact phrases people use. They become the headline of your landing page on day 5.

Paul Graham suggests one more test. Ask who wants this so much that they would use a rough first version from a tiny startup they've never heard of. If you can't name those people, keep looking.

Which signals mean the startup idea is worth building?

A startup idea is worth building when several signals point the same way. One strong interview is a story. Five interviews that describe the same problem in the same words are a pattern.

Look for these, from weakest to strongest:

• Search interest for the problem holds steady or grows.
• Competitors or workarounds exist, and people complain about them.
• Interviewees describe the problem before you mention it, and they've tried to fix it.
• People leave their email on your one-page offer or ask when it's ready.
• People pay a deposit or a pre-order before the product exists.

Graham's advice is to find a small group who want the product badly rather than a big group who want it a little. A narrow, deep need is easier to win and easier to grow from.

Which AI tools help you validate a startup idea faster?

AI tools can now help with every step of validation, so one founder can run the nine checks without a team. Each step has its own kind of tool:

• Desk research: AI research assistants like Perplexity and Claude search the web and summarize what they find, and Perplexity cites its sources.
• Customer interviews: AI interviewers like Listen Labs, Outset, and AskEngine run voice or text interviews and group what people say. Listen Labs can recruit from its own global panel and interview in more than 120 languages.
• Surveys: Tally makes free forms that work like a doc, and Typeform asks one question at a time.
• Landing pages: Framer and Lovable turn a description into a page you can share the same day.
• Pre-orders: Stripe payment links take money without a checkout you have to build.

Keep a person in the loop. Let the agents gather and sort the evidence, and make the final call yourself.

How does ZeroCrew help you validate a startup idea?

ZeroCrew helps you get from idea to evidence by showing which tools founders search for and use for each step. The AI customer interview tools page ranks interviewers by how many people in the US search for each one by name, and the survey and feedback tools page does the same for forms and feature boards. The AI research assistants page adds which tools founders call worth the price.

Every tool page has a free check that asks whether the tool is right for what you're building. Describe your idea, and Agent Zero, ZeroCrew's AI assistant, compares the tool with its site, what founders say about it, and its competitors. You get a yes, a no, or a better tool.

Your own AI agent can search the ZeroCrew directory too. Connect the public MCP server at https://zerocrew.ai/api/mcp/public to Claude, ChatGPT, or Cursor, with no sign-in, and ask it for the best tool for each check while you plan your week.

If your idea is a tool for founders, ZeroCrew members can test it for you. A ZeroCrew Operator, who wins a seat in the Friday auction, can post a bounty: a small task with a reward, open to every member, with up to 50 places and 7 days to take part by default. That fits a one-week validation. When the evidence says build, launch on ZeroCrew and read how to become a maker.

What should you do when the checks say no?

A "no" from validation is useful information, not a failure. The 2024 replication found that founders trained to test their ideas ended more ideas, and that they changed strategy a few times rather than never or constantly. That's what a good search looks like.

When an idea fails a check, look at which one. A weak search signal with strong interviews may mean people don't have words for the problem yet. Strong interest with no pre-orders often means the price, the buyer, or the offer is wrong. Change one thing, then run the week again.

Which idea will you test this week?

Pick the idea where you already know the customer. Write the one sentence today, book the first three conversations, and find the agent for each check in the ZeroCrew directory. In seven days, you can have real evidence to decide with.