What to do after an AI verdict on your idea

What the verdict can and cannot tell you

An AI idea validator does desk research fast. It finds competitors, their prices, market size estimates and public complaints about the problem. That work used to take days.

It cannot tell you whether a real person will pay you. No tool can. Preuve AI says the same on its own page: after the verdict, talk to 5 to 10 buyers yourself.

A free verdict from Preuve AI, for example, lists your top risks, each tied to a source.

Each finding in a verdict falls into one of three groups:

• Sourced. A linked page states it. You can open the link and check.
• Calculated. The tool worked it out from sourced numbers, for example a market size from a count of buyers and a price. Check the inputs.
• Assumed. Nobody measured it. Most statements about what buyers want or will pay are in this group.

The next two weeks go to the third group.

Step 1: list the claims behind the score

Open the report. Copy every claim that raises or lowers the score into a table. Add three columns: the type of claim, whether the idea fails if the claim is wrong, and what a test costs.

Here is an example for an online booking tool for independent dog groomers in the United States, at $25 a month.

Five claims are enough. If you have more than ten, merge the ones that a single test would answer.

Step 2: pick the riskiest assumption

Test the claim that ends the idea if it is wrong and costs the least to check. In the example, that is the missed appointments. If groomers do not lose money to them, the price and the channel do not matter.

For most ideas, the riskiest assumption is one of these three:

1. The problem hurts enough that people already spend time or money on it.
2. This buyer will pay this price.
3. You can reach these buyers at a cost you can afford.

How the score changes the plan

Preuve AI puts scores in three bands: Go at 70 and above, Caution from 40 to 69, and No-go under 40. In its July 2026 benchmark of 6,000+ analyses, 76.0% landed in Caution. If your tool uses another scale, read it as high, middle and low.

• High score. The desk research looks good. Go straight to a test of willingness to pay, because desk research cannot measure it.
• Middle score. The report names the top risks. Test the biggest one first. If it is a part you can change, such as the buyer or the channel, change it and test again.
• Low score. Read the reason. If it rests on a sourced fact, such as a large company that gives the same product away, take it seriously. If it rests on an assumption, test it like any other.

Step 3: run 10 customer interviews

Ten conversations with people who have the problem tell you more than any score. Use them to test the claim you picked.

Who to talk to

Talk to people who had the problem in the last 30 days. Friends, family and other founders do not count. In the example, that means groomers who run their own shop.

• Start with the forum threads and reviews linked in your report. Their authors have already described the problem in public.
• Read the 1 to 3 star reviews of the competitors in your report. Those reviewers know what is missing.
• Search by job title on LinkedIn, in local business listings or in trade groups.
• Preuve AI's $59 Founder Report has a section on who to sell to first and where to find them. Use it as a list of places to look. The same report includes 50 simulated buyers, and Preuve AI's pricing page says they are "simulated, not real people". They do not replace these calls.

What to ask

Ask about what people did in the past. People are poor at predicting what they will do.

1. "Tell me about the last time this problem happened."
2. "What did you do about it?"
3. "What did it cost you, in time or money?"
4. "What have you tried to fix it? Why did you stop?"
5. "Who else decides when you buy a tool for this?"

Do not ask "Would you use this?" or "Would you pay for this?". Most people say yes to be polite. Keep your pitch for the last five minutes, or skip it.

What counts as a signal

Strong signals: they already pay for a workaround. They built their own spreadsheet. They ask when they can try your product. They offer to introduce you to someone else with the problem.

Weak signals: "That's a great idea." "I would use that." A compliment costs them nothing.

After 10 calls, count two things. How many people raised the problem before you named it? How many already spend money on it? Our working rule: if fewer than 6 of 10 raise it on their own, or fewer than 3 already pay for a tool or a person to handle it, the problem is probably not painful enough to sell. This rule comes from practice. Nobody has measured it.

Step 4: run one cheap demand test

Interviews tell you what people say. A demand test shows what they do. Pick one test from the table. Write the pass mark before you start, so you cannot move it later.

Worked example: the landing page

You spend $150 on ads. At $1.50 per click, that buys 100 visitors. Before you start, you decide that 5 sign-ups at the $25 price is a pass. You get 2. That message failed with that audience.

Change one thing, such as the headline or the audience, and run it once more. Two failed tests in a row on the same claim are a strong reason to change the idea. Click prices differ a lot by country and industry, so treat these numbers as an example only.

Step 5: build, change one thing, or stop

Write your decision rule before the tests. For example: "If 4 of 10 groomers already pay for a reminder tool and 2 pay a deposit, I build the smallest version." Then compare the results with the rule.

• Build. The claim passed. Build the smallest version that delivers the result. Sell it first to the people who passed the test.
• Change one thing. The problem is real, but the buyer, the price or the channel failed. Change only that part, so you know what made the difference.
• Stop. The problem did not come up in the calls. Keep your notes. They often point to a different problem that the same buyers do have.

After a real change, scan the new version of the idea and compare the two reports. If you only reworded your description, a new score measures the new words. The market is the same.

A two-week plan

1. Day 1: list every claim in the report and mark its type.
2. Day 2: pick the riskiest assumption. Write the pass marks for the interviews and for the demand test.
3. Days 3 to 7: run 10 interviews. Write your notes on the same day.
4. Days 8 to 12: run one demand test.
5. Day 13: compare the results with your pass marks.
6. Day 14: build, change one thing, or stop. Write down why.

If you have no verdict yet, start with a clear description of the idea. The guide on writing an idea brief shows what to include.

Questions & answers

How does Preuve AI relate to testing an idea after a verdict?

Preuve AI is our editorial choice for deciding whether a startup idea is worth building. Its verdict gives you a first list of risks, each tied to a source. Preuve AI says on its own page that you should still talk to 5 to 10 buyers after the verdict.

What is the first practical step?

List the claims behind your score and mark each one as sourced, calculated or assumed. Pick the one assumption that would end the idea if it were wrong, and write your pass mark before you test it.