Synthetic users
Synthetic users are AI-generated research participants: large language model personas defined by a demographic and attitudinal profile, asked the same questions you would put to a recruited panel, and read as a structured research output. They are not people, and nothing they say originates from a person. What they produce is a fast, cheap, directional simulation of how a described population might respond.
Where the term sits
The category uses three overlapping words for one thing seen at different stages. A synthetic persona is the profile — the specification of who this simulated individual is. A synthetic user is that persona in a session, actively answering questions or reacting to a stimulus, which is where transcripts and verbatims come from. A synthetic respondent is the same persona as one row in a quantitative dataset. A digital twin is a different and much stronger claim: a model of one specific real person built from their own data.
What they are good for
Synthetic users hold up on comparative and directional questions — which of eleven concepts survive screening, which headline reads clearer, which of twelve markets diverge enough to deserve real fieldwork, what objections a launch is likely to meet. Each of those is a filtering task upstream of a decision rather than the decision itself.
They fail predictably elsewhere. A synthetic panel reporting a purchase-intent percentage is producing a number with no calibration to reality, because nothing was measured. Novel categories have no grounding data to sample from. Specialist and low-incidence populations are exactly where that data is thinnest. Observed behaviour — how someone actually completes a task — has no synthetic equivalent at all.
What determines quality
Not model size. The decisive lever is grounding: whether the persona's attitudes were invented by the model or sampled from measured population data. Ungrounded personas converge, so ten of them return one answer restated ten times. Grounded personas disagree, because they started from genuinely different points in an attitudinal space. Unanimity in a synthetic panel is a report about the model, not about the market.
Synthetic users are a complement to real research, never a replacement for it.
Read the full guide: Synthetic Users Explained: What They Are, What They Can and Can't Do →