I stopped asking AI to “validate” startup ideas and started breaking them into assumptions

While building an AI system for startup analysis, I ran into a problem with the usual workflow.

The prompt is often something like:

“Is this a good startup idea?”

The problem is that this produces one big conclusion from a bunch of smaller assumptions.

So I changed the workflow.

Instead of asking AI for a verdict, I break the idea into separate questions:

Problem

• Who actually experiences it?

• How often does it happen?

• What happens if they do nothing?

Current solution

• What are people doing today?

• Is the workaround free, manual, or already good enough?

Business

• Who pays?

• Why would they pay?

• What would make them switch?

Product

• What has to be true for the product to work?

• Which feature is actually essential?

Risk

• What assumption, if false, makes the whole idea weaker?

That changed the output quite a bit.

Instead of ending with:

“This is a promising idea.”

you can end with:

“These are the three assumptions you still haven't tested.”

I think that's a much more useful job for AI in the early stages of a startup.

The AI doesn't need to make the decision.

It needs to help you understand what the decision actually depends on.

I'm curious whether other founders use a similar approach, or whether you still prefer getting one overall assessment of an idea.