Simulated people first

Simulated people are AI‑generated participants the Sage Engine builds from a participant model you define. They give you a hypothesis about a design or a question, not evidence. Say “simulated people,” not “virtual people.” Virtual sounds made up; these responses are still data you can check.

Why start here

If the simulated panel already falls over on weak wording or a first click that lands in the wrong place, you should not spend recruiting time confirming a mess. Change the flow or the claim, clone the study if needed, and run again. That pass is still a hypothesis, not a verdict.

AI Panel runs draw from your monthly credits. Most methods cost 1 credit per run. Prototype Test costs 2. Results come back labeled AI Panel, and they never sit in the same pile as real respondents.

How the Sage Engine differs from a one‑off chat role‑play

You can ask an AI agent in a chat tool to pretend to be a user. That can be useful for a quick opinion. It is usually a one‑off improvisation from a short persona prompt, and changing the wording can change the answer.

User Sage’s AI Panel works from a structured model of each simulated person: role, experience, confidence, goals, constraints, behavioral tendencies, and what they have already seen in the study. Those stay consistent across the method, tasks, and questions. You get individual responses and a summary you can save, share, compare, and rerun. It is still a hypothesis. When a decision matters, run the same study with Your Panel or GenPop.

AI Panel never uses real participant PII to generate simulated responses. Real participant responses are never used to train or improve AI.

Run the panel and read Pulse

1. From your drafted study, run the AI Panel when you are ready to spend the credit.
2. Read individual simulated responses, not only the summary. User Sage does not blend people into a single average.
3. Check Pulse on the AI path: Good, Watch, or Needs work. On AI Panel, Pulse flags whether simulated people look varied enough for a first pass, or whether they may sound too alike or too eager to please. Filling in product context helps.
4. Write down what you would check next with real people. That list is the point of the hypothesis.

What not to claim

Do not put an AI Panel percentage in a deck as proof about your users. If simulated people clicked Continue first at some rate, that is a claim about those simulated people. It is a place to look, not a ship decision. The product keeps AI and real results in separate Findings columns so the meeting cannot quietly treat them as the same.

There are no accuracy guarantees for AI panels in this guide, and there should not be any in your write‑up either.

When to stop at the hypothesis

Low‑stakes copy tweaks, early exploration, and “does this task even make sense” checks can stop at a strong simulated pass plus your product judgment. When the decision is expensive to reverse, tied to trust, or about to ship to many people, go and see (chapter 6).