Check answer quality with tests

With administrative Chat permissions, open Settings → Administration → Insights → Tests. Additional evaluation views appear only when enabled.

Prepare a test

Create or choose a question collection. Use questions that represent the real use case and add suitable reference answers when supported. Keep the initial conditions comparable.

Run the test

Select the offered agent, evaluation models and options. Start the run and wait for completion. Open its results and review individual answers, errors and available ratings.

Interpret results

Compare runs only when the collection and configuration are appropriate. An aggregate score does not replace checking sources, factual correctness and unintended actions. Questions without a reference answer cannot provide the same reference-based assessment as questions with one.

Changes to the agent or its sources may affect results. Record the configuration and test time for internal quality review.