Measuring visibility in generative search

A brand mention, an article link and a customer enquiry are different outcomes. A single AI visibility score should not replace them. We first define the questions and markets that matter to the business and what the research needs to establish.

Define an observation

For each trial, we record the question, language, tool, date and conditions. We distinguish the company name appearing, a link being provided and the description being accurate. A consistent question set makes later comparisons more useful.

Separate a sample from traffic

A manual review of answers shows a selected part of the user experience. It cannot provide a complete count of all impressions. Reports should disclose sample size, missing results and method changes instead of presenting a sample as the whole market.

Use traffic data carefully

Traffic from Google's AI features is included in the Web search type in Search Console. Reporting should separate what the available data shows from assumptions about why a change occurred. We do not attribute an entire increase to one answer format.

Documentation: Google Search Central.

Example: a focused study

For B2B services, we select questions about a problem, comparisons and choosing a provider. We record whether answers describe the offer accurately and where links lead. The quality of enquiries received during the period is assessed separately.

Make the report useful for decisions

Results should identify missing information, inaccurate descriptions and pages worth improving. We retain observations where the brand is absent too. The team can then understand the limits and repeat the study using the same method.