Synthetic User Testing
Synthetic user testing uses AI-generated simulated users — driven by an LLM prompted to behave like a specific persona — to interact with a product the way a real user would, generating realistic usage paths and edge cases at a volume and speed no human test team could match, before or alongside real user testing.
Traditional usability and exploratory testing is bottlenecked by human time — a handful of real testers can only generate so many session paths in a day. Synthetic users, each an LLM prompted with a persona and a goal, such as a frustrated first-time user trying to cancel a subscription, can run hundreds of varied sessions against a staging environment in the same time, surfacing usage paths a fixed test script wouldn't have scripted.
The value is in variety and volume, not depth — synthetic users are good at generating a wide spread of plausible interaction paths and flagging where the product's flow breaks or confuses, but they're a poor substitute for genuine human judgment on subjective quality, since the "user" being simulated has no real stake in the outcome.
The main pitfall is treating synthetic user results as equivalent to real user research — a synthetic user's behavior reflects the LLM's model of how a persona would behave, which can miss the specific, often irrational ways real humans actually use software, so it's best used to widen test coverage cheaply, with real user testing still validating the conclusions that matter most.
