Synthetic user testing in a real product

StrangerTest belongs to B.

Ask a simulated respondent whether your onboarding is clear and you get a plausible answer about onboarding in general. Send a simulated user through your onboarding and you find out whether it reached the first useful screen, where it hesitated, and what it clicked instead of the thing you meant it to click.

The second kind of result can be checked. There is a browser record: the pages, the actions, the screens before and after. That is what makes it possible to tell a real product problem from a mistake by the agent — which matters, because agents make them.

• ·Whether real customers want the product or will pay for it.
• ·Your conversion rate, or any number that stands in for it.
• ·How people with real stakes, real data and real feelings will experience it.
• ·Anything about a part of the product the test could not safely reach — which a good report names rather than skips.

Simulated users are not customers, and should never be presented as if they were. They are a way to see your product without the knowledge you cannot switch off.

The broader category is covered in AI user testing: what it can — and cannot — tell you. How StrangerTest’s users are kept cold, and how their accounts are checked, is in How StrangerTest works.