“AI visibility” is an umbrella term, not one standardized score. A useful report states which output it measures, what counts in the numerator and denominator, which answers were sampled, and when they were collected.
Brand mentions
A brand mention is a response that names the brand at least once. For a defined sample, calculate mention rate as responses with a mention divided by valid responses reviewed. If the same response names the brand four times, decide whether your unit is “response with a mention” or total mention instances and label it. Vendor definitions can differ.
Citations and cited sources
A visible citation is a link or source reference shown in the answer. Report citations to your domain separately from citations to any source that mentions your brand. Count answers with at least one owned-domain citation separately from the number of citation links and unique cited URLs. A brand can be mentioned without its site being cited, or its site can be cited without the brand name appearing in the prose.
Position and prominence
Position describes where a brand or link appears in a particular answer. It is not a universal ranking comparable with a Google blue-link position. Define the extraction rule: order of brand mentions, order of linked sources, or a platform-specific score. Do not average values from different answer formats unless the method is defensible.
Share of voice
Share of voice should name its comparison set. One possible formula is: a brand’s qualifying mentions divided by all qualifying brand mentions across a defined competitor set and prompt sample. Another vendor may use cited answers, topics, or weighted search demand. Label the formula and do not present a proprietary score as market share.
Prompt coverage and visibility scores
Prompt coverage is the portion of a named prompt set where a brand appears or is cited. A vendor’s visibility score may combine several factors, such as breadth across topics and frequency within a prompt set. Read the provider’s own definition before comparing scores. Ahrefs and Semrush, for example, document provider-specific metric systems; their outputs are not a common standard.
Impressions and referrals
Google Search Console’s Generative AI performance report reports impressions for supported Google Search generative features. Bing Webmaster Tools’ AI Performance report describes citation counts and cited pages for its supported AI experiences. An impression, citation, and AI-referred website session each measure a different stage. Analytics referrals capture visits, not answers that were seen without a click.
Crawler activity
Crawler requests measure access attempts or successful fetches in logs. They do not prove that content was indexed, used in an answer, or cited. Keep bot requests separate from answer visibility and referral traffic.
A minimum reporting template
For each metric, record: definition; numerator; denominator; platform and answer mode; prompt/source set; market and language; collection method; collection date and cadence; valid/failed runs; and exclusions. Show raw counts beside percentages so readers can assess small samples. If a provider does not disclose the denominator, say so rather than reverse-engineering a universal formula.
For a repeatable protocol, see how to measure AI visibility. For product-specific calculations, read the Ahrefs definitions, Semrush metrics, and the current Bing AI Performance documentation.
What citation studies can tell us about measurement
Early studies show why reports must name the outcome and sample. A comparison of 55,936 queries found that 37% of domains were unique to the LLM-based search systems in that study. A separate public-service study found at least one flagged source in 65 of 187 reviewed web-search answers. These are different outcomes from different settings, not a universal AI visibility rate.
Use studies like these to sharpen metric definitions and sampling—not as cross-platform benchmarks. See the source-selection evidence and its limits.