Define what counts
A mention is text identifying the brand. A citation is a visible link to a source. A correct answer describes the business accurately, even if the brand is not linked. A referral is a website visit attributed to a source, subject to analytics limitations.
These are different signals. A mention without a link is not referral traffic, and a citation to a third-party page is not a visit to your own site.
Build a representative prompt set
Choose perhaps 10 to 20 questions that reflect real purchasing or support tasks. Include branded and non-branded prompts, and note the intended audience, language and region. The number is a sampling design choice, not a Google-prescribed minimum.
For an agency, examples might include “When is AEO useful?” and “What should be checked before rebuilding a website?” Avoid rewriting prompts after seeing the desired answer, as that biases comparisons.
Record every observation consistently
For each test, record the exact prompt, platform or product, date, location/language where known, answer text or a permitted excerpt, whether the brand appears, whether a visible source link exists, the linked domain and whether the description is accurate.
Use a stable cadence and separate manual checks from automated ones. Respect platform policies and do not assume an automated interface reproduces the experience of every end user.
Use a transparent reporting table
Prompt coverage: share of tested prompts for which the brand was mentioned in that specific sample.
Linked-source frequency: share of tested responses containing a visible citation to a specified domain.
Factual accuracy: share of checked claims about your organization that match verified source facts.
Always disclose the prompt set, number of observations, date range and limitations. These metrics are defined for your internal experiment; they are not platform-wide ranking scores.
Connect visibility to business outcomes carefully
Track real clicks and conversions using search analytics and website events where possible. AI platform referrer labels are not always stable or comprehensive, and some visits cannot be attributed reliably.
Use Search Console for Google web-search performance and indexing diagnostics, not as proof of every appearance or citation in every generative answer. Report unknown attribution as unknown rather than filling gaps with estimates.
A monthly review process
Keep the prompt set stable for the review period. Repeat tests under recorded conditions, investigate incorrect claims and note changes in source pages or relevant documentation. Record one concrete action per finding, such as correcting an outdated service description or adding evidence to an underexplained guide.
Prioritize clarity and correctness. More mentions are not useful if the answers misrepresent what a company offers.
A practical AI visibility measurement worksheet
Start with a fixed question list and record individual observations rather than inventing a “ChatGPT rank.” The table below is a blank template, not ThreeZero performance data.
| Date | Platform | Exact prompt | Brand mentioned? | Linked citation? | Source URL | Factually accurate? |
|---|---|---|---|---|---|---|
| YYYY-MM-DD | Platform + mode | Paste exact question | Yes / No | Yes / No | Observed URL or none | Yes / No / Unknown |
| YYYY-MM-DD | Platform + mode | Second question | Yes / No | Yes / No | Observed URL or none | Yes / No / Unknown |
Mentions per tested response = responses containing the brand name ÷ total responses tested. Linked citations per tested response = responses linking to your domain ÷ total responses tested. Report the denominator, platform, date, language and whether the sample was collected manually. These are your own sample metrics, not platform-wide ranking statistics.
CSV header you can reuse
date,platform,mode,language,prompt,brand_mentioned,linked_citation,source_url,claim_accuracy,notesDo not merge multiple platforms' outputs into a single score without explaining sampling differences. Keep unknown answers as unknown, and measure actual enquiries and visits separately.
Sources and further reading
These references support the search-engine guidance in this article. Practical examples are explanatory, not claims of client outcomes.
Questions readers ask
Can AI visibility be measured with a single ranking?
No universal AI ranking measures all questions, platforms and sessions. Report clearly defined prompt-based observations and source links, including the sample and limitations.
Do AI search citations always produce website visits?
No. A cited link may not be clicked, and traffic attribution may be incomplete. Measure on-site referrals separately from mentions and citations.