An AI search baseline needs a fixed sample before it needs a headline metric. Decide which questions and conditions you will repeat so later observations have a meaningful comparison.
What to record
- Choose questions covering discovery, comparison, and product constraints; keep each group separate.
- Fix the interface, locale, and prompt wording for the comparison.
- Choose and record a repeat count before seeing favorable or unfavorable answers.
- Preserve all attempts, including missing citations and unsupported claims.
Illustrative plan: five questions run three times create fifteen observations. If four cite your page, report four of fifteen in that sample—not a universal visibility rate. A new interface or prompt set starts a different comparison.
A check you can run
Write the sampling plan before the first run. Save the exact question list, conditions, repeat count, and review date. Keep it unchanged through the comparison or explain the difference explicitly.
Download preview image