Website checks and AI business facts
A test log for incorrect business facts in AI answers
A test log needs the exact prompt, platform, date and observed response before choosing a correction. “AI says something wrong” is too vague to retest. Separate observed output from hypotheses about its source.
Understand the task
AI answers can vary by run, location and context. Correcting your own page does not guarantee an external system will use it next time. The log makes work reproducible and identifies controllable sources rather than promising citations or rankings.
Prepare these inputs
- Exact prompt and relevant non-sensitive conversation context.
- Platform, date and incorrect wording.
- Authoritative source containing the correct fact and its owner.
A practical workflow
- Save the observation without private customer conversations. Record whether a source citation actually appeared.
- Check owned pages and relevant managed profiles. Correct documented inconsistencies in sources you control.
- Repeat the agreed question and record changed and unchanged outcomes. Do not generalise one response across platforms.
Worked example, not a customer case
An illustrative example
Illustrative log entry: “Platform X, specified date, opening-hours prompt, answer states old hours; owned page shows new hours.” Investigate the source hypothesis separately. Without a displayed citation, the log must not claim a particular directory supplied the answer.
Test failures and exceptions
| Test input | Expected outcome |
|---|---|
| A new answer is correct on a different platform. | Record it separately from the original test. |
| The prompt changes during retesting. | Record the change rather than calling it the same test. |
Evaluate whether it helps
Track specific factual errors, displayed citations and repeat outcomes. Explain sample size and variation. Accuracy within a small log is not a universal AI-visibility score or guarantee.
Next step: Start with one fact and one reproducible prompt, keeping correction and retest separate.
Sources and scope
Examples are editorial exercises using fictional information. They do not describe completed customer tests or activated integrations. The authority source below provides AI and data background rather than approval of the exercise.
Want to clarify your own workflow?
Bring one task and a non-sensitive example. Scope, price and timing are agreed individually before work. Reading a guide activates no customer messages or system actions.
Describe your task