Test a Figma prototype with AI personas

How Figma prototype testing works with AI personas in Versive, from connecting your file to reading the results.

Figma prototype testing with AI personas imports selected frames into Versive and has one or more AI users review them in a chosen order. The personas react frame by frame through either an interview-style conversation or a structured heuristic review. Results include transcripts, screenshots, and prioritized recommendations, usually within minutes and without participant recruitment.

Use it for an early review before a flow is built, preparation for a design review, or a comparison of two screen versions. Do not use it as the sole evidence for a high-stakes decision. The workflow is to connect the prototype, configure the test, select personas and a mode, then inspect the results.

Connect your Figma prototype

Versive imports a prototype one of two ways.

The first is pasting a Figma URL directly. Versive reads the file and imports its frames as screens automatically. It does not execute the interaction graph defined in Figma, so after import you choose the frames to include and arrange the sequence the AI will review. The file needs link sharing turned on, "Anyone with the link can view," or the import will not be able to read it.

The second is the Figma plugin, listed in the Figma Community as "Versive: Instant Usability Tests." It lets you select the specific pages and frames you want to test from inside Figma itself, authenticate with your Versive account over OAuth, and create the test without switching tabs. You can reconnect that account any time.

After import, choose which frames to include and their order. A test of ten focused screens makes findings easier to attribute to a specific step, while forty screens from several flows can blur the source of a problem. A narrow, single-flow test also costs less to rerun as the design changes.

A Figma prototype test lives inside a project, the same organizing unit that holds image tests and website tests. Projects move through draft, active, and completed or archived stages, and a project moves to active the first time you run a test in it. You can duplicate a project to test a new version of the same design later, and every run keeps a snapshot of the exact screens and configuration it tested, so old results stay meaningful even after you have iterated on the design.

Set up the test

Before you run anything, a few settings shape what the AI actually looks for.

Write the goal as a research brief. The goal field provides the context you would give a moderator before a session: what the design is, who it serves, and what you want to learn. "First-time user setting up a workspace; we want to know if the invite step is discoverable" is more specific than "test my onboarding flow" and directs attention to the step under study.

Pick a test mode. Prototype tests run in one of two modes:

• Conversational is the default. An AI moderator interviews the persona about each screen, producing a transcript that covers first impressions, expectations, confusion, and next actions. Use it to investigate the reasons behind a usability problem in a format closer to a moderated session.
• Expert audit runs a single AI expert against a set of usability heuristics, Nielsen's ten by default, rating each one Good, Fair, Poor, or Not Applicable and listing the specific issues found per screen. You can replace or add custom heuristics, including your own design system rules. Use it for a fast, systematic QA pass rather than simulated reactions, such as before a design review or as a pre-launch checklist. Heuristic evaluation with AI: a working method walks through running that mode well.

Run the same screens through both modes in one project when you need persona reactions and a systematic audit.

Set the test sections. These are the dimensions each screen gets evaluated on. Versive defaults to Initial Impressions, User Actions and Behavior, Usability Issues, and Recommendations, and you can edit or replace them to match what you are actually studying.

Add per-screen notes and scripts. You can attach notes to individual screens, and in conversational mode, a custom interview question for a specific screen, useful when there is one step you need every persona to weigh in on regardless of how the conversation is otherwise going.

Turn on a SUS questionnaire, optionally. Every persona can fill out a System Usability Scale questionnaire after its run, producing a standard 0 to 100 score per run that is comparable across design iterations. Treat those scores relatively, comparing version A to version B, rather than as an absolute benchmark against human-normed data.

Consider "ignore content accuracy." If your prototype still has placeholder text or dummy data, this setting tells the AI to focus on structure and flow instead of reacting to the lorem ipsum as if it were real copy.

Pick personas and run the test

A test needs at least one AI persona. Add reusable personas from the library, either written manually or distilled from research data, or write one inline for a single test. Each persona has its own execution count. For example, you can compare five executions of a "busy first-time user" with three of a "returning power user."

Each persona execution is one test run and typically completes in minutes. Prototype and image tests use one credit per persona execution. Credits are charged only when a run succeeds, including a partially completed prototype run in which some screens finished. A failed run can be retried once at no charge unless the retry succeeds.

After changing the design, rerun the same personas in the same project. Comparing rounds shows whether the finding changed and provides more context than either raw score alone.

Read the results

Every run produces its own detailed record. In conversational mode, that is a full interview transcript per screen, moderator and persona turns, alongside the screen image. In expert audit mode, it is heuristic-by-heuristic ratings with the specific issues found. If you turned on SUS, you also get that persona's questionnaire answers and computed score.

Above the individual runs, Versive builds an aggregated test summary across every persona and execution: an overall results rating from Excellent down to Critical with an explanation, key takeaways tagged positive, negative, or neutral, and prioritized recommendations linked back to the specific screens, runs, and quotes that support them. Summaries generate on demand and can be regenerated once you have added more runs.

Look for repetition across personas and executions before acting. An issue that appears in four of five runs has more support than a one-off. Treat priority labels P0, P1, and P2 as the AI's severity assessment, then check the highest priorities against the underlying evidence and your judgment. You can share findings through a public, read-only result link that requires no Versive account, or export a report to Word or Markdown. See Results & reports for the contents of each run and summary.

When an AI test fits, and when you need real users

AI prototype tests can identify apparent points of confusion quickly enough to run after design changes. They suit early concept checks, usability review before a design critique, comparisons between two versions of a flow, and design QA between rounds of user research.

Treat the findings as hypotheses ranked by likelihood. Confirm findings that affect a launch decision with a small study using real participants. What are synthetic users, and when should you trust them? explains that limitation, and Moderated vs. unmoderated usability testing compares formats for the follow-up study.

Once a flow has actually shipped, you can point an AI test at the live site instead of a prototype. See Run a usability test on a live website for testing the real thing rather than a click-through mockup.

Connect a Figma file to a new project and run one focused flow with one or two personas. Review the source evidence in the report before expanding the test.