Metrics That Matter in AI Search, and the Ones to Ignore

The problem with AI visibility numbers

Dashboards for AI search often show a score, a rank or a share of voice. Before you trust one, ask what it measured. The SparkToro study from January 2026 asked 600 volunteers to run 12 prompts through ChatGPT, Claude and Google's AI a combined 2,961 times. The chance that two responses contained the same list of brands was under 1 in 100, and the chance of the same list in the same order was under 1 in 1,000. The authors suggest measuring visibility as a percentage over many prompts run many times, and they advise distrusting any claim of a "ranking position" in AI.

That does not make measurement useless. It means the right measures are rates over many runs, plus the data platforms actually provide, plus real business outcomes.

Metric 1: Google Search Console

Google says traffic from AI Overviews and AI Mode is counted in the Performance report under the Web search type. It does not break them out. Watch impressions, clicks and click-through rate for your key pages over time, and annotate the dates you publish or change pages. Pew's research is a reminder of why click-through can fall even when impressions hold: users clicked a traditional result in 8% of visits to pages with an AI summary and 15% of visits to pages without one, and about 18% of the searches in the sample had a summary.

Metric 2: Bing Webmaster Tools AI Performance

Microsoft's AI Performance report, in public preview, shows when your site is cited in AI answers across Microsoft Copilot, AI-generated summaries in Bing and select partner integrations. The metrics include total citations, average cited pages, grounding queries and page-level citation activity. Microsoft states it does not report placement within an answer, ranking, authority or page importance, and it counts citations, not clicks. It is the closest thing to a first-party AI citation report that we know of.

Metric 3: AI referral visits in GA4

Some AI tools send visitors who click a link. In Google Analytics 4 you can create a custom channel group that collects those sources, which lets you track them as one channel. Google's help page describes custom channel groups and includes an example for AI assistants. Treat the total as a floor: some visits arrive without a recognizable referrer, so referral counts undercount. Our guide to tracking AI referral traffic in GA4 walks through the setup.

Metric 4: Appearance rate across repeated prompts

This is the measure closest to what clients experience. Write a fixed list of prompts a client might ask, run each one several times, and record whether your firm or pages are named or cited. Your appearance rate is the number of runs where you appear divided by total runs. Because results vary, use enough runs that one lucky or unlucky answer does not drive the number. The AI answer audit gives the full method.

Metric 5: Intake source on real inquiries

The metric that matters most is whether people contact the firm. Add "How did you hear about us?" to intake, with options that include AI assistants and search. Record the answer in your case management or CRM. It will not capture everyone, but it can catch clients who saw your name in an AI answer, then searched for you directly. That behavior does not show up as an AI referral.

Metrics to ignore or discount

• A single "position in AI" number. The evidence on answer variability makes it unreliable.
• Screenshots of one answer. One run says little.
• Scores with no public method. Ask what was measured, how many prompts and runs, and when. Some vendors publish a rubric. GrowthAgents, one of this site's sponsors, publishes one on its methodology page, which you can read as an example of what a stated method looks like. Judge any rubric by whether its sample and method are open to inspection.
• Top-10 rankings as a proxy. Ahrefs found that about 38% of URLs cited in AI Overviews also ranked in the top 10, down from about 76% in its earlier study, so rankings alone do not predict citation.

A simple monthly dashboard