How to measure AI visibility: mentions, citations and recommendations

Define the observation before creating a score

Use three separate labels. A mention means the answer names the business. A citation means it links to a relevant source page. A recommendation means the answer suggests considering the business in a stated context. One response can satisfy several labels, and a citation to an educational article may contain no business recommendation.

Keep the wording as evidence. A sentence describing a company as an example differs from a sentence recommending it for a particular need. Automated text checks can help triage observations, but a reviewer should inspect ambiguous language.

Keep the comparison conditions visible

Choose benchmark questions that represent actual services and customer decisions. Include relevant location or requirements naturally. Avoid inserting your brand into every question if the goal is to understand unaided discovery. Record provider, model or interface, timestamp and whether live research was used.

A developer API sample and a consumer chat session are different test conditions. Keep them in separate series. When comparing competitors, use the same prompt set and record missing observations as missing; an unavailable provider is not a confirmed absence from its answers.

Use platform reporting where it is available

Bing’s AI Performance reporting provides citation activity and cited-page information across supported Microsoft experiences. Its grounding-query view is a sample with a defined scope. It should not be presented as a complete record of everything people ask across all chatbots.

Google Search Console provides impressions, clicks and query-based views for the site’s search performance. Website analytics can then help assess visits and agreed conversion events. Treat those records as complementary measurements rather than combining them into an unexplained universal AI ranking.

Connect each finding to a useful next action

If a competing source is cited, inspect what decision its page helps a reader make. Your response might be a clearer comparison, an original implementation example or a correction to stale service information. Copying the competitor’s wording is not an improvement plan.

Report the observation window and counts alongside the outcome: pages published, sources cited, referral visits and qualified inquiries. A change in one sampled answer can guide investigation, but it cannot establish that a content edit caused every later result.

1. Save the original answer and referenced URLs.
2. Classify mentions, linked citations and recommendation wording.
3. Compare against the previous matching observation window.
4. Assign an evidence-based update with an owner.
5. Recheck with consistent questions after the update is published.

Common questions

Sources & further reading

• Bing: AI Performance reporting
• Google: Search Console performance reporting
• xAI: encountered sources and inline citations

Read our editorial policy