Build a small test set
Choose questions that represent real customer needs. Include common product questions, compatibility conditions, policy exceptions and questions the business cannot answer from its current knowledge.
Write the approved answer or expected fallback beside each test question. This gives you something more concrete to evaluate than whether the response sounds friendly.
Run the private preview
- Select the website and save the latest bot setup.
- Open Relay’s private preview.
- Ask the test questions using ordinary customer wording.
- Try a shorter or ambiguous version of a few questions.
- Compare each answer with the approved facts.
- Record gaps and update the appropriate Brain entry.
Do not fix every answer by adding more tone instructions. A factual gap belongs in Q&A, product knowledge or another suitable source record. A wording problem may belong in the bot’s greeting, tone or fallback.
Check boundaries
Ask something outside the business knowledge. The bot should give an appropriate limitation or next step instead of inventing a policy.
Check that it does not promise unsupported actions such as booking a service, refunding an order or transferring to a human. A link to an approved contact destination can be useful, but the bot’s text should describe what the customer must actually do.
Repeat after meaningful changes
Changing an active policy, product record or bot setup can affect answers. Retest the questions most likely to be affected. A previously successful answer does not establish that all later revisions behave identically.
Once public chat is enabled, review visitor conversations and Brain recommendations. Private preview is a useful controlled check; real visitor questions can expose phrasing and knowledge gaps that your initial test set did not cover.