Synthetic Personas: Keep an Evidence Ledger, Not a Fictional Biography
Make synthetic personas useful by linking each meaningful trait to source evidence or an explicit assumption, then reviewing contradictions and decision relevance.
Synthetic personas are generated profiles intended to represent a possible user context. They become more useful when each decision-relevant trait is linked to evidence or clearly marked as an assumption. A vivid biography is not evidence, and an interactive persona is not the person whose experiences informed it.
This guide explains how to maintain a persona's evidence ledger after the initial profile is created. For initial scope and stereotype checks, see our synthetic personas design guide.
Separate the profile from its supporting material
A profile might say a team manager reviews status updates weekly and cannot authorize a new vendor. Those constraints affect how the manager interprets a product page. By contrast, an invented hobby or childhood story usually does not help answer the research question and may create false confidence in the persona's realism.
Keep source material outside the generated narrative. Record the trait, its source, the interpretation and its uncertainty. If a trait came from a researcher hypothesis, say so. Do not let a model rewrite that hypothesis as a customer quote during later analysis.
NN/g defines synthetic users as AI-generated profiles that attempt to mimic user groups and cautions that these generated findings are not a replacement for studying people. That distinction applies even when a profile was inspired by real research. 1
Build a ledger that survives edits
Use stable descriptions for relevant constraints rather than continually expanding the persona. A teammate should be able to tell whether an edited profile changes the research context. If a visitor gains purchasing authority between comparisons, the comparison no longer holds that condition constant.
This hypothetical ledger demonstrates the format. It contains no actual customer data or Cairo study results.
Trace the trait to its basis
Illustrative fields for an operations-manager profile.
| Trait | Basis | Allowed interpretation |
|---|---|---|
| Reviews weekly updates | Hypothetical research theme | Explore whether the page explains review |
| Cannot approve a vendor | Explicit study assumption | Separate interest from authority |
| Uses an existing workflow | Scenario condition | Ask about switching effort |
| Values automatic accuracy | Unverified assumption | Do not present as a customer finding |
A ledger need not be a complex database. A concise document can work if it stays attached to the study version. The important property is traceability: a generated answer should not become more authoritative merely because it has been copied into a polished persona card.
Keep contradictory evidence visible
Real research rarely describes one perfectly consistent customer archetype. One manager may welcome automation; another may prefer manual review. Do not erase the disagreement by writing that the persona “wants automation but values control” without stating the contexts in which each preference appeared.
Where the evidence supports different contexts, use separate scenario profiles. Where the evidence is incomplete, preserve the unknown. If a source concerns one organization or workflow, avoid generalizing it to an entire demographic group.
A synthetic persona may produce an answer that conflicts with its ledger. For example, it may enthusiastically approve a purchase despite lacking authority. Inspect the decision and restate the task constraints rather than treating the generated answer as a newly discovered behavior.
Write task context without leaking the desired answer
A neutral task might ask the manager to decide whether the product deserves further investigation. A leading task might ask how the product would improve the manager's productivity. The second assumes an improvement before the persona evaluates the page.
Give the persona the source page and necessary prior knowledge, not a private explanation of every benefit. If the page does not explain the review workflow, the generated visitor should not know it just because the study brief does. Otherwise you are testing the prompt rather than the product experience.
Ask for specific unresolved questions. Short, concrete responses are easier to inspect than elaborate speeches. Do not quote a synthetic persona in marketing as though it were a customer testimonial.
Move from persona context to product action
Cairo uses segmented synthetic profiles to explore public pages and possible reactions. Use the ledger's constraints when setting the audience brief, then inspect whether the journey and recommendation fit that context. A proposed page edit should answer a real information gap, not invent capabilities to satisfy a persona.
If a meaningful constraint rests only on an assumption, the next action may be a human study instead of a redesign. GOV.UK's moderated usability guidance offers a practical reference for observing people attempt realistic tasks. 2
Keep any customer material permissioned and minimized. A synthetic profile is not automatically anonymous if it reproduces identifiable details from a source. Review what you share with any service and what you publish in reports.
Questions about synthetic personas
Are evidence-based synthetic personas real participants?
No. Source evidence can improve context, but subsequent generated answers remain simulated and model-dependent.
Should I add personality details to make responses realistic?
Only when relevant and supported. More detail can introduce stereotypes or assumptions without improving the decision you are studying.
When should a persona be retired?
Revisit it when its source context is outdated, the task changes, or contradictory evidence invalidates its assumptions. For panel-level checks, use the synthetic audience coverage audit.