Record which experience you observed and whether an AI answer actually appeared.
Name the experience
An AI Overview on a search results page, an AI Mode interaction, and a separate assistant experience are different observations. Record the product, question, market, language, time, and any accessible contextual details. Do not combine them solely because they share a provider.
Distinguish ‘no AI answer appeared’ from a failed collection and from an AI answer that omitted your brand. These are different states. A measurement that collapses all three into zero visibility hides the underlying behavior.
Build a baseline you can inspect
Use customer questions across discovery, comparison, and validation. Save the complete visible answer and links where available. Record collection limits honestly, including features or states your method cannot reach.
Inspect actual linked pages rather than inferring a source from the domain name. Check whether the page supports the nearby claim and whether it is current. An answer can use a product page, community discussion, or local record for different purposes.
Separate access from representation
Check whether relevant pages can be reached and contain accurate, visible information. Then compare how that information is represented in answers. A page being accessible is a prerequisite you can verify; a particular recommendation is an outcome you observe.
Keep technical work aligned with the documented requirements of the product being tested. Avoid an ‘AI optimization’ change that creates contradictory content or undermines an established search and publishing process.
Report by task and surface
Show results by question group and experience, with sample sizes and example answers. A blended average may be useful as a summary only if the reader can inspect its components and weighting.
Combine the answer audit with customer and analytics evidence. If local questions produce incorrect hours, fix the records. If comparisons omit an important constraint, clarify the relevant page. The next action should follow the finding rather than a generic platform checklist.
