AI search analytics: where the data lives and how to read it
By Vouch7 min read

Until recently, AI search was a measurement black box: buyers asked assistants for recommendations, and brands had no first-party data about what happened next. That changed in 2026. Google and Microsoft now both ship analytics for their AI surfaces, and together with your web analytics and sampled visibility tracking they form a workable measurement stack. This piece maps the four sources: what each one shows, what each one misses, and how to read them together.
The four questions AI search analytics must answer
Every useful number in this space answers one of four questions:
- Is my content being used in AI answers? Impressions and citations on Google's and Microsoft's AI surfaces.
- Does AI search send me traffic? Referral visits from assistant domains.
- Can AI systems reach my content at all? Crawler access, the leading indicator under everything else.
- What do assistants say about me when no link is involved? Mentions, sentiment and recommendations inside generated answers.
No single tool answers all four. Here is what each source contributes, as of November 2026.
Source 1: Google Search Console's generative AI insights
Google announced new insights for website owners in Search Console on June 3, 2026, and the post confirms they rolled out to all websites worldwide by August 31, 2026 (New opportunities, control and insights for website owners). The insights cover your pages' appearance in Google's generative AI features and include:
- Impressions metrics for generative AI Search features
- Which of your pages appear in AI responses
- The countries where those appearances happen
Google says it will introduce additional metrics over time. The same announcement added a Search Console control letting sites opt out of appearing in and grounding generative AI features such as AI Overviews and AI Mode; Google states that sites that opt out receive no traffic or impressions from those features, and that the control is not a ranking signal for search outside them.
The scale behind these numbers: Google reported AI Overviews at over 2.5 billion monthly active users and AI Mode past one billion monthly users in June 2026, per the same post.
What it does not give you today: the phrases that triggered your appearances, citation counts in the Bing sense, and anything about non-Google assistants. Treat it as your Google-specific impressions layer.
Source 2: Bing Webmaster Tools' AI Performance report
Microsoft launched AI Performance in Bing Webmaster Tools in public preview on February 10, 2026 (Introducing AI Performance in Bing Webmaster Tools). It shows how your content appears across Microsoft Copilot, AI-generated summaries in Bing and select partner integrations. Its metrics:
- Total Citations: how often your content was displayed as a source in AI-generated answers over the selected period.
- Average Cited Pages: the average number of unique pages from your site cited per day.
- Grounding queries: the key phrases the AI used when retrieving content that ended up referenced in answers. Microsoft notes this is a sample of overall citation activity.
- Page-level citation activity: citation counts for individual URLs, so you can see which pages get referenced most.
Microsoft is explicit about what citations do not mean: they do not indicate placement, ranking or importance within any single answer. Bing also recommends IndexNow to keep content fresh for AI experiences, since current content matters for citation.
Note what Bing gives that Google currently does not: query-level grounding phrases. If you want to know which questions your cited pages are answering in AI surfaces, Bing's report is the only first-party source of the two, as of November 2026.
Source 3: AI referral traffic in GA4
AI answers carry links, and clicks on those links arrive in your analytics as referrals from assistant domains such as chatgpt.com, perplexity.ai and copilot.microsoft.com. Segmenting those referrers in GA4 gives you the traffic half of the picture: which assistants send visits, to which pages, and what those visitors do next. The full setup is in our guide to seeing AI referral traffic in GA4.
Keep two limits in mind when reading these numbers. First, referral data can only capture visits where the referrer survives the click; some app and privacy contexts strip it, so treat the figures as a floor rather than a total. Second, and more importantly, an AI answer can shape a buying decision without any click at all. A buyer who reads a comparison in an assistant and then types your URL directly shows up in no AI report. That gap is why the fourth source exists.
Source 4: Sampled visibility across the assistants
The two consoles share a structural limit: each covers its own company's AI surfaces. For a cross-assistant view, and for everything that happens inside answers without a citation or click, the established method is sampled visibility tracking. You run a fixed set of buying questions through the assistants on a schedule and record the patterns: mention rate, citation rate, survival across the follow-up conversation, sentiment and recommendation rate. Our guide to AI search visibility defines those metrics, and the best AI visibility tools in 2026 comparison covers the tools that automate the sampling.
This is the layer where you learn that an assistant names a rival at the objection stage, or that your pricing page is cited while your positioning is misquoted. Neither console can tell you that, because neither sees the whole answer.
Reading the four sources together
A few working principles. These are editorial recommendations, not documented rules:
- Keep the sources separate. They count different things in different units. A Bing citation, a Search Console impression and a GA4 session are three different events; never add them into one made-up total.
- Watch trends within each source. A sustained fall in Bing citations or Search Console AI impressions is a signal even when you cannot reconcile the absolute numbers across tools.
- Check access before blaming content. If citations drop, verify that AI crawlers can still reach the affected pages before rewriting anything. Our free AI crawler checker tests that in seconds.
- Let queries guide content. Bing's grounding phrases show what cited pages are being used for; pages that are indexed but rarely cited are your improvement candidates, as Microsoft's own guidance suggests.
- Correlate, carefully. A rise in AI impressions followed by a rise in AI referrals is consistent with visibility turning into traffic, but the two datasets measure different things, so treat the link as a hypothesis to keep testing, not proof.
How Vouch helps
Vouch covers the fourth layer, the part the consoles cannot see. It runs your prompts daily across seven AI assistants (ChatGPT, Claude, Gemini, Perplexity, Copilot, Google AI Overviews and Google AI Mode), and its Threads follow a six-question buying conversation from Discovery to Decision on ChatGPT, Claude, Gemini and Perplexity, recording where your brand drops out and who replaces it. Mention, survival and recommendation rates roll into the weekly Vouch Rate, a 0 to 100 score where sentiment moves the result by at most 0.1.
Pair that with the two consoles and your GA4 segments and you have all four questions answered: Google's and Microsoft's first-party numbers for their surfaces, referral data for the clicks, and Vouch for the answers themselves. Vouch is invite-only for now; join the Vouch waitlist and every invitation starts with a 14-day free trial.


