Measuring AI visibility
Keep citations, referral visits and business results distinct.
AI visibility is too broad to measure as one number without a definition. A displayed source link, a visit from an AI product and a useful business action describe different events. Decide which question you need to answer before collecting data. A small observation log can be useful, provided its limits stay attached to the result.
Record citation observations
For each observation, keep the exact question, product and mode, date, language, visible answer and source URLs. Include relevant conditions such as whether you were signed in or continuing an existing conversation. Use a defined set of questions and repeat the procedure consistently. Preserve cases where no source is shown as well as cases where your page appears.
The result describes what you observed under those conditions. It is not a universal ranking, a market-share estimate or evidence that every user sees the same answer. Google says its AI features can use different models and techniques, with varying responses and links, in its AI features documentation. Do not attribute a changed answer to your page edit without considering other explanations.
Keep visits in a separate record
A citation can be seen without being clicked. A referral record describes a visit attributed to a source, subject to the information your reporting can see. Missing or incomplete attribution means you cannot reconstruct every journey from a referrer alone. This site does not install analytics; a site owner considering measurement should choose an appropriate, privacy-conscious method for their own context.
For Google’s AI features, Google says appearances are included within overall Search Console performance for the Web search type. Its Search Console introduction describes search performance reporting. Do not label all Google organic traffic “AI traffic” or assume a general Search Console total isolates a particular generated answer.
Define a useful outcome
Hypothetical example: a pottery studio wants qualified class inquiries. A citation is one observation, a visitor opening the class page is another, and an inquiry asking about an available class is a third. Track the relevant events separately if you have an appropriate measurement method. A rise in one does not automatically establish a rise in the others.
Even when visits and inquiries increase after an edit, seasonality, a promotion or other site changes may offer alternative explanations. Keep a dated change log and compare comparable periods where possible. Small samples and several simultaneous changes usually limit causal interpretation. Report that uncertainty instead of turning a sequence of events into a proven effect.
Use observations to choose the next check
If source links repeatedly misrepresent a condition on your page, inspect whether that condition is clear in its original context. If users visit but cannot complete the next step, review the page and journey. If nothing changes, record the no-change result. Avoid inventing a positive story simply because work was completed.
A useful report separates actions, observations and interpretations. State which pages changed, what the observation procedure captured, what traffic attribution can establish and which business outcome matters. End with the next decision the evidence supports. A completed checklist or a count of citations should never be presented as a probability of commercial success.