Synthetic user
What it does
A synthetic user is a customer profile a model can speak as. Grounding decides whether it's worth anything: each trait traces to something a real person said in a real session, and the profile carries that citation wherever it goes. A profile spun up from a market description answers just as fluently with nothing behind it, which is the failure mode the technique is known for. The idea also travels under synthetic participant, AI persona, evidence-based persona and, in market research, digital twin.
This skill builds profiles from your research repository. It clusters sessions that behave alike, writes the profile from what the cluster actually contains, and footnotes each attribute to the session and moment it came from. Where the evidence stops, it says so and logs the gap as a question for your next round. Once a profile exists you can hand it an artifact, a PRD, a concept, a design or a draft survey question, and get a reaction written from that evidence base, citations still attached so you can check whether the reaction was earned.
What you get back
- One profile per cluster, listing the sessions it was built from by name and date.
- Per-attribute citations: the session, the moment and the line of transcript behind each claim.
- A gap list covering what the profile cannot speak to, written as questions for your next study.
- Reactions to any artifact you paste in, each traced to supporting evidence or marked as unsupported.
- A strength note per profile: how many sessions it rests on and how old the newest one is.
How to use it
Example Prompts:
"Build synthetic users from our onboarding research from the last two quarters."
"Have the finance lead profile read this PRD and tell me which parts it has no evidence about."
"Which of these profiles is resting on the thinnest evidence?"
What it won't do
It won't stand in for talking to real people, and nothing it produces should be reported as a finding. A profile can only recombine what your repository already holds, so it will never surprise you the way a real interview does, and it knows nothing about what changed after your last session. Use it to sharpen what you'll ask before you spend recruiting budget.
Questions
Are synthetic users a replacement for interviews?
No. A profile is a compressed reading of research you already ran, so it can only repeat what real people already told you. Anything you need to learn fresh, and anything you would defend in front of a stakeholder, still means talking to real people.
What stops it inventing things?
Every attribute has to point at a session, and anything without one is marked as unsupported rather than written in. When a profile is asked something outside its evidence, it says the evidence doesn't cover it and adds the question to the gap list.
How much research do I need before profiles are useful?
Enough sessions in one cluster that a pattern is visible instead of a single person's story. The skill reports the cluster size behind each profile and flags the thin ones, so you can see when a profile is really one interview wearing a persona name.
