Test Your AI-Drafted Idea With Real People in ChatGPT Before You Build It
A pre-revenue founder can run a first real-person check on an AI-drafted concept from inside an AI client, but it should test the riskiest assumption, not prove willingness to pay.
You have an AI-drafted concept, a rough prototype, and no revenue. The next move is a small real-person check that can stop the idea before you spend weeks building it.
If you ask an AI client to validate a startup idea, it will hand you a plan. The missing piece is a real person: evidence before building, not praise.
Run the check from inside your AI client
On September 24, 2026, UserTesting announced that its Model Context Protocol servers for the UserTesting and User Interviews platforms are available.
The new UserTesting and User Interviews MCP servers are in Early Access and let teams find participants, set up studies, and check AI-created concepts inside ChatGPT, Claude, Figma Make, Gemini, and other Model Context Protocol-compatible apps, backed by a participant pool of over seven million people in 34 countries.
The User Interviews MCP integration works with any AI tool that supports Model Context Protocol, including Claude, ChatGPT, and Cursor.
It functions as a full research process where an AI assistant sets up projects, recruits participants, and surfaces session data.
In User Interviews, project management, team settings, and analysis tools are still only available in the app rather than through MCP.
The User Interviews MCP integration does not use customer data to train third-party AI models and uses OAuth authentication tied to the organization's existing permissions.
For a pre-revenue founder, the useful move is to treat the MCP servers as a customer validation tool. You can talk to potential customers before you build a product or hire a research team. The AI client handles setup, and real participants supply the evidence.
- Name the riskiest assumption.
- Set pass/fail before recruiting.
- Choose a usability test or an interview.
- Run the MCP study from ChatGPT or Claude.
- Review sessions for evidence.
- Record what it does not answer, especially willingness to pay.
Test the riskiest assumption before the build
Name the riskiest assumption in one sentence: a person in your target group will use this because of a specific pain, not because the idea sounds smart. Say that pain plainly, or you are not ready to test.
Say you have a tool that helps people schedule calls. The riskiest assumption may be that busy professionals will let it handle their calendar. When the real risk is whether they trust it with their time, test that, not the onboarding screen.
Set pass/fail before recruiting. Decide what evidence would make you stop, pivot, or continue. A useful threshold is behavioral: the assumption survives on a participant completing the core action without your help, and fails on a long explanation. Write that down where you will see it after the sessions. Fix the written threshold and plan through the first session.
Choose the format that matches the risk. A usability test suits an unclear core action. An interview suits a problem you have not yet named. Pick one clear format before recruiting: interviews for that early problem, usability tests for a rough flow that needs a break point. Keep the conversation focused; a long, unfocused mix of formats will muddy the evidence.
Run a user study from the AI client. Ask it to recruit test users, create the study, and surface session data. Keep the task small: a single core action, a clear participant profile, and a clear question. You want a study plan, not a conclusion or a pitch. Start with one study, not a research program.
Treat polite interest as a signal, not proof of demand. People will be kind to a founder, and that kindness is not purchase evidence. Vague enthusiasm is not enough; keep looking.
Turn sessions into an honest log
Listen for what people do, not what they say they might do. Note where they hesitate, ask for clarification, or abandon the task. A short clip of confusion is more useful than a long compliment.
Keep notes tied to the assumption. A useful note is that the participant hesitated at the permission step. A less useful note is that they seemed excited. Point to the moment, or discard the note.
Record what the test does not answer. A short session can show confusion or interest. It cannot prove willingness to pay, pricing, or long-term retention. Put that limitation next to the result so your next decision stays honest.
Put the result in a simple log: assumption, evidence, decision, and what remains unknown. That log is the artifact you will show a future cofounder, investor, or future self when the idea starts to sound better than the evidence.