Measure visibility in AI answers responsibly

Our method

Define a relevant question set, record the platform, date, language and answer context, and check whether the cited page actually supports the answer. Track referral traffic and qualified enquiries separately from the observation log.

How to assess progress

Account for answer variability and limited attribution. Avoid presenting hand-picked screenshots as universal performance. Search crawling, model training and user-triggered browsing are different mechanisms.

Decision

Decide whether you need to observe brand recognition, source citation or commercially useful traffic. These are different questions and require separate records. Define the relevant market, language and buyer situation before collecting prompts, so a rise in mentions does not automatically become a claim of business growth.

Deliverable

Deliver an observation ledger with platform, account context, date, language, question category, mention status, citation destination and factual issue. Keep a separate traffic table with identifiable referrals and completed actions. Explain the denominator whenever reporting a mention rate and retain the original evidence for each observation.

Pitfall

Changing the question panel while reporting a continuous percentage creates a misleading trend. Version the panel and compare unchanged questions separately when adding coverage. An uncited mention cannot prove which page influenced the answer, and a referral does not establish that every earlier answer displayed the same information.

Read Google AI impressions separately

Search Console’s Generative AI report records impressions of links in AI Overviews and AI Mode. Compare impressions by page, country, device and date; dates use Pacific Time. These impressions overlap the Web report, so do not add the totals. The documented report provides neither AI clicks nor search queries: missing data is not zero. Google: Generative AI performance report

Check Settings → Search generative AI as well. Inclusion makes participation possible; it does not guarantee display or control model training. Google: Search generative AI control

Keep ChatGPT citations in the observation sheet below. Measure visits, qualified enquiries and revenue separately, using your analytics and business records. Google impressions alone cannot establish those outcomes.

Track Bing citations separately

In Bing Webmaster Tools, AI Performance covers Microsoft Copilot, AI answers in Bing and selected partner experiences—not every AI service. Follow total citations, citations by URL and the daily average of unique cited pages over the same reporting period.

Grounding queries are phrases used to retrieve supporting content, not necessarily the user’s exact question. Their coverage is sampled. Citation Share describes your share of citations for a particular grounding query; it is neither a ranking nor a share of traffic.

When available, Topics Portfolio (beta) groups queries into themes. Its estimates and automated classifications are directional; limited activity can leave topics or the feature absent. Keep these observations separate from Google’s report and from measured visits and qualified enquiries. A citation is not a click or a conversion, and improvements do not guarantee future citations.