# Synthetic Persona or User Observation? Choose the Right Evidence for the Decision

Use synthetic personas to rehearse hypotheses, not stand in for user evidence. Match field observation to decisions about real behavior.

- Canonical URL: https://www.harmate.com/en/blog/synthetic-persona-or-user-observation-choose-the-right-evidence-for-the-decision
- Author: Harmate Team
- Published: 2026-11-15
- Updated: 2026-11-15T01:50:04.546525+00:00
- Language: en

## Content

# Synthetic Persona or User Observation? Choose the Right Evidence for the Decision

A synthetic persona can help a team rehearse a prompt or imagine a possible path through a product. It cannot establish what actual users do, understand, need, or prefer. Choose the method by asking what evidence the next product decision requires.

## Match the method to the question

Start with the decision, not the tool. If you are checking task wording, a rehearsal may expose assumptions. To learn how people move through a flow, notice a status, or abandon a task, observe people who fit the intended context using the product or prototype.

A synthetic output can be useful as a hypothesis: “Someone might look for a draft list.” That statement is not a participant observation. It does not tell you whether anyone actually looks for a list, how often, or why. Nielsen Norman Group’s discussion of synthetic users is a useful reminder to keep generated perspectives separate from evidence about people. [Read about synthetic users and research limits](https://www.nngroup.com/articles/synthetic-users/).

## Decision example: returning to a saved questionnaire

A team is deciding whether to redesign the “save and return later” flow in a six-page questionnaire. The product question is whether first-time users can leave midway and resume with their answers intact.

A synthetic profile can help the team rehearse the scenario, uncover missing details, and list paths worth testing. It might suggest that a participant will search the navigation for a saved draft. The team records this as a **test hypothesis**, not a finding.

The decision depends on actual behavior in the interface. Recruit people who match the study’s context, show the same build and starting state, and give a neutral task: “You need to stop now and continue tomorrow. Show me what you would do.” Record the action sequence and ask neutral follow-ups such as “What are you looking for?” The team can then report what occurred in each session and what remains uncertain.

| Item | Label it as | Can support | Cannot support by itself |
|---|---|---|---|
| “Search for a draft list” from a generated profile | Synthetic hypothesis | A path to include in a test | That users take this path |
| A participant opens a menu and returns to the form | Session observation | What happened in that session | A population rate or universal pattern |
| A participant says a status label reassures them | Participant explanation | How that person interpreted the cue | That the cue works for everyone |

## Use the synthetic pass as preparation

Write down the synthetic rehearsal’s source and purpose. Keep generated suggestions in a hypothesis column and turn useful ones into observable questions. Do not blend them into participant notes or count them as users. A generated persona does not replace recruiting participants.

For a real user study, specify eligibility, the interface state, the task, the recording method, and how you will preserve context. If the decision concerns needs or observed behavior, collect evidence from people in that context. If the decision concerns whether your own prompt or test plan is coherent, a rehearsal may be enough to improve the instrument—but it does not answer the user-research question.

## Make the decision traceable

Keep five fields: product decision, evidence needed, source, what it supports, and what remains unknown. This prevents a plausible synthetic answer from silently becoming a customer claim and shows when fieldwork is necessary.

The example above is a controlled illustration, not a study of a real questionnaire or its users. It demonstrates how to choose evidence; it does not show that synthetic personas predict user behavior or that any particular interface is usable.

Use the [evidence-based buyer persona guide](https://www.harmate.com/en/blog/buyer-persona-build-it-from-evidence-not-assumptions) to work from interviews and observed behaviour. For a software check with a fixed fictional input, see the [synthetic persona agent test](https://www.harmate.com/en/blog/how-to-test-an-ai-agent-with-a-synthetic-persona).