Brand mentions, recommendations, and citations need different reporting

An answer can mention a brand without recommending it. It can recommend a product under a condition. It can cite a page whose publisher is not one of the recommended vendors. Those are distinct observations, and a report becomes misleading when it turns them into one unexplained visibility number.

Choose the measurement question first. If you want to know whether the brand is discussed accurately, mentions and context matter. If you want to understand selection, inspect recommendations and their conditions. If you want to know which owned resources are used as sources, inspect citations and their exact destinations.

A mention means the answer names the brand or an unambiguous equivalent. A recommendation means the answer presents it as a relevant choice for the user's task, with any conditions retained. A citation means the interface attributes supporting material to a linked source. These are proposed coding definitions for your study; platform reports may use different definitions.

Allow overlap. One answer can mention, recommend, and cite the same business. Another can mention it critically while citing an unrelated publisher. A report should preserve those combinations rather than force every answer into a single favorable bucket.

Add an ambiguous status. Reviewers may disagree about whether a neutral list counts as a recommendation. Keep the disputed record and write a rule for future cases. A small calibration review is more useful than silently letting each analyst interpret the category differently.

Imagine a hypothetical panel of 20 answers. Eight mention the brand, three recommend it, and five cite an owned page. Two of the recommendations also cite an owned page. The counts describe different properties of the same 20 answers. Adding eight, three, and five does not mean the brand appeared in 16 distinct answers.

Report the denominator and overlap. You could say “eight of 20 answers mentioned the brand; three of 20 recommended it; five of 20 cited an owned page.” Then explain whether the recommendations were conditional and whether the citations supported the relevant product claim. These numbers are illustrative, not a benchmark or an observed result for StackM.

Now consider quality. An owned page could be cited to support a criticism or a limitation. Counting it as a citation is appropriate; calling it a positive recommendation is not. Preserve the actual context so the team can choose a useful follow-up.

Google's June 3, 2026 announcement, updated August 31, describes dedicated generative-AI performance views with impressions and related dimensions, with the data also included in overall performance reporting. That does not make those impressions equivalent to the manually coded answers in your panel. See the dated Google update for the current stated scope.

Bing describes Citation Share as an observational share of citations for a grounding query. Its announcement explicitly distinguishes it from traffic share and content quality. Keep that metric labeled according to Bing's definition, rather than renaming it market share or combining it with another engine's measure.

A practical dashboard can show these sources side by side, with separate definitions and time windows. It should not total them into one number merely because they all concern AI visibility. Different systems observe different events and may cover different populations.

An inaccurate mention suggests a claim or identity investigation. A recommendation with a recurring condition may reveal a product-fit question your own page should explain. A citation to an obsolete resource suggests reviewing that destination and its replacement. The observation points toward an investigation; it does not establish why the system produced the answer.

Track identifiable referral behavior and qualified enquiries separately where the available analytics support it. Do not assume a citation produced a visit or that a visit produced a sale. Self-reported discovery can add useful context, but it is another view of the journey and may overlap with directly observed activity.

Include the question set, observation dates, environment, category definitions, denominator, and material changes to the study. If the panel changed, show the stable subset separately. Otherwise, an apparent gain may simply reflect the addition of easier branded questions.

Use the evidence log to preserve individual records and the visibility measurement guide to plan the study. The strongest report makes its limits easy to understand and still gives the team a specific next decision.

For the next implementation step, use Calibrate two reviewers coding AI recommendations.

For the next implementation step, use Resolve an ambiguous brand mention before counting it.