Synthetic Personas
Run research every single day with unlimited personas grounded in your real customer data — a thousand, ten thousand, a million at once. About 80% as accurate as a real user, at a scale no research team could ever reach.
Real research is slow.
Synthetic research is daily.
Talking to real users is irreplaceable — but you only ever hear from a sliver of your audience, and it takes weeks.
AI isn’t a user, and it’s never going to replace hearing it from the horse’s mouth.
But when you can run research every day against thousands of personas grounded in your actual customer data, see consensus, and read the reasoning — that’s a different kind of signal. Directionally, it’s very, very good.
Consensus on copy.
Feedback on UI.
Put options in front of a panel and see which one lands. Or paste a UI screenshot and get annotated reactions back. Both return statistical analysis across every variant.
“Which of these five would you pick?”
Paste a pricing page, a prototype, a Figma export — get annotated reactions.
1,000× the research.
10 points less accurate.
Reality sits at 100% — the ceiling you’ll never fully hit. The best user research gets you ~90%, but only every few weeks. Synthetic personas land at ~80% — and you can run them a thousand times a week.
Lower accuracy per study — dramatically higher signal per month.
And every one of them is grounded in your real customer data.
About 80% as accurate
as a real person
In our own evals on BuildBetter, the last 20% is the stuff we can’t know — a sudden priority change, a relational shift, a goal that moved this morning. Everything else? Surprisingly directional.
Sentiment, reasoning, consensus, objections, priorities by segment, and how a message lands across archetypes.
Immediate things — a goal change this week, a relationship shift, a new incident. Use synthetic research alongside real calls, not instead of them.
Not “pretend you’re a PM.”
Every persona is deeply detailed.
Each persona carries goals, needs, motivations, behaviors, preferences, role, company archetype, and context. Every question they answer is filtered through that full picture — not a sentence of role-play.
Personas don’t exist
in a vacuum
Every persona sits inside a synthetic company — with its own stack, stage, team shape, and buying process. Same depth, applied to the org.
Talk at a persona,
or with one
Interview style is up to you. Open the persona one-on-one, or build a panel and run a study.
Talk at the persona
Speak in first person, as if you're running the interview. The persona responds in character.
Talk with the persona
Use a mediator between you and the persona — closer to a moderated session with an observer.
Stack archetypes on archetypes
Open the matrix builder, select your personas, and cross them against company archetypes. In a few clicks you can generate 140+ variants — same title, different goals; same company stage, different security posture. Every variable becomes a lever.
Not pulled from thin air
Every study pulls in real signals as evidence. People have already reacted to this kind of thing on calls, in tickets, in Slack. That context is what sources consensus — not a vibe check from a language model.
Top positive reactions
See the three strongest positive responses by persona and by segment.
Top negative reactions
Surface the sharpest objections and the specific quotes behind them.
Verdict + reasoning
Read the summary and the reasoning, then jump straight into the persona to push back.
What teams are asking
synthetic personas today
A rolling look at the studies running against synthetic personas right now. Each one grounded in your real customer data.
“Would Series B PMs get confused by our new pricing page?”
Run sentiment across product-manager archetypes before you ship. Catch the objections that would have cost you a deal.
Run user research every day...
