Synthetic Users: What They Are and How to Choose Their Role

Synthetic users are AI-generated representations of people that produce simulated responses or explore defined experiences under specified assumptions. They can help a team develop questions and examine possible interpretations. They cannot establish what real customers need, how often a problem occurs, or whether people will buy.

The selection question is therefore not simply which tool to use. It is which uncertainty you want to explore, what the simulation can inspect, and what evidence must come from people. Choose that role before generating a panel or reading a persuasive report.

Separate a simulated user from a real participant

A synthetic user combines a model, profile instructions, supplied context, and a task. Its output reflects those ingredients rather than a person's lived experience. A detailed biography does not change that distinction. Neither does a first-person explanation that sounds emotionally convincing.

A real participant can reveal an unexpected constraint grounded in an actual situation. A simulated profile can propose a constraint, but the team must investigate whether it exists. Treat the difference as an evidence boundary, not a small disclaimer at the bottom of a presentation.

Nielsen Norman Group's guidance describes synthetic users as potentially useful for desk research and hypothesis generation while emphasizing that they cannot replace research with real people. Its evaluation also describes overly favorable responses and shallow accounts of needs. That supports a narrow exploratory role, not generated validation of demand.

For input terminology, see synthetic persona design. A profile specifies assumptions; a simulated response does not verify them.

Choose the method by the question

Begin with a decision you could make differently after the exercise. “Understand our users” is too broad. “Identify unclear language to investigate before recruiting buyers” gives the simulation a bounded job.

Distinguish questions about the material from questions about the market. A website can contain an unexplained term regardless of how many visitors notice it. A claim that the term causes abandonment requires evidence beyond simulation. A generated objection can inspire an interview question without proving that objection is common.

Do not use synthetic users as a workaround when recruiting is difficult but the decision still depends on human experience. Difficulty obtaining evidence does not lower the standard required for a consequential claim.

Inspect the setup, not the apparent personality

Evaluate an approach through its inputs and boundaries. Can you identify the supplied profile, task, material, and limits of the environment? Can you distinguish visible content from model inference? Can another reviewer understand how an interpretation entered the report?

In Cairo, exploration can begin with a URL or idea and use demographic, socioeconomic, and psychometric synthetic profiles. Website exploration is bounded, and audience views and reports organize generated material. These features define a way to inspect assumptions; they do not make the panel a representative sample.

Cairo supports declarative changes to copies that preserve the original and same-panel variant comparisons. This is not a statistical A/B test. A simulated journey also does not establish a real purchase or submission.

Use Cairo and Rosetta for the available workflow. For technical evaluation criteria, use the existing framework for evaluating synthetic audiences, rather than treating the checklist here as a model benchmark.

Hypothetical example: select a role for a pricing page

Hypothetical example, not a reported result: A scheduling-software team wants to revise a pricing page. Its question is whether “active workspace” communicates what gets billed. It has no evidence yet that this phrase causes purchase hesitation.

The team chooses bounded exploration rather than asking generated users whether they would subscribe. It supplies the current page and defines two contexts: someone comparing tools for a small team and someone reviewing costs across several locations. These are assumptions, not discovered customer segments.

The task is to locate the billing unit, explain its meaning using visible information, and identify what remains unspecified. The researcher records page text separately from generated interpretation. If an output assumes each location requires another subscription, the researcher checks whether the page supports that inference.

Next, the team drafts a clarification while preserving the original. A same-panel comparison can surface differences in simulated interpretation. It cannot demonstrate fewer abandoned purchases or establish which wording real buyers prefer.

The deliverable is a brief: the ambiguous term, competing interpretations, proposed clarification, and questions for actual purchasing stakeholders. The simulation has helped define an investigation, not completed it.

Use a decision checklist before starting

• Name the specific uncertainty and decision it informs.
• Explain why simulated interpretation is relevant.
• Identify the source or assumption behind each profile constraint.
• Define what can be inspected and where exploration must stop.
• Separate generated explanations from visible page content.
• Assign human follow-up for consequential behavioral claims.
• Decide who reviews contradictory or unsupported outputs.

Stop when the exercise produces a reviewable set of hypotheses. Generating more profiles is not automatically the next useful step.

Know when not to select synthetic users

Do not use a simulated panel to estimate demand, establish accessibility, measure conversion, or certify safety. Higher-stakes decisions deserve stronger validation, not more elaborate fictional biographies.

Synthetic responses can share model-level assumptions even when profiles differ. Agreement is not independent confirmation. Fluent detail can conceal missing context, and plausible objections can distract from problems real people would raise first.

Select synthetic users when output remains a labeled input to thinking: a candidate interpretation, interview topic, or journey question. The synthetic user research guide explains how to preserve that boundary from question to claim.

Next step: Explore your public page or product idea in Cairo, define one question and an audience, and use the simulated report to plan what to investigate next.