Google and Bing Now Report AI Visibility. Here Is What Their Dashboards Still Cannot Tell You.

AI visibility reporting just became first-party data

In February 2026, Microsoft introduced an AI Performance report in Bing Webmaster Tools. It shows total citations, cited pages, sampled grounding queries, and citation trends across Microsoft Copilot, AI-generated Bing summaries, and selected partner experiences. In June, Google began testing dedicated Search Generative AI performance reports in Search Console, separating impressions in AI Overviews, AI Mode, and generative Discover experiences from the broader performance view.

For GEO and AI SEO, this is a real change. Publishers no longer have to infer every AI appearance from referral logs or manually collected screenshots. Google can show when one of your URLs appeared in its generative surfaces. Bing can show when your content was displayed as a source and which phrases its systems used for grounding.

The launch also confirms something the market spent two years arguing about: generative visibility is distinct enough to deserve its own reporting. The catch is that each dashboard observes a different part of the journey.

Google measures appearances; Bing measures citations

Google’s early report is page-centric. It includes impressions, pages, countries, devices, and dates. An impression means a URL from the site appeared in a generative feature. That is useful exposure data, but it does not say whether the URL supplied the decisive claim, where it appeared in the response, or whether the business was one of the options recommended.

Bing’s report is citation-centric. A total citation is a displayed source reference. Its grounding-query sample is especially useful because it begins to reveal the retrieval language behind the answer rather than only the user’s visible prompt. Microsoft explicitly cautions that these metrics do not indicate placement, authority, page importance, or the role a page played in an individual answer.

Neither product is defective for drawing that boundary. First-party webmaster tools can report only what their own surfaces observe. The mistake is relabeling either number as total “AI visibility” across ChatGPT, Claude, Gemini, Perplexity, Copilot, and every answer format.

A citation is not a recommendation

Consider an answer to “Who is a good estate-planning attorney in Manhattan for a cross-border family?” The assistant might recommend three firms, cite a state court page for probate rules, and cite a fourth law firm’s explainer for a tax point. That fourth firm earned a citation but not a place on the shortlist. One of the recommended firms might be named from the assistant’s existing knowledge without an owned-domain citation at all.

This is why Viclaro separates mentions from citations. Citation reporting answers “was this page used or displayed as a source?” Recommendation monitoring answers “was this business offered to the buyer?” Those events can overlap, but neither implies the other.

For a publisher, citation volume may be the primary outcome. For a service firm, it is usually an intermediate one. The commercial outcome begins with recommendation presence, recommendation strength, and position in the shortlist—not merely a footnote somewhere in the response.

Use a three-layer AI search scorecard

The cleanest measurement stack has three layers. First is platform exposure: Google generative impressions, Bing citations, cited URLs, and grounding queries. Second is market presence: recommendation rate, share of voice, shortlist position, and model coverage across a frozen panel of buyer prompts. Third is business impact: detectable AI referrals, qualified inquiries, booked consultations, and revenue.

Do not add those layers into one synthetic score. Reconcile them. A page whose Google AI impressions rise while the firm’s recommendation rate stays flat is earning exposure without improving shortlist presence. A firm whose recommendation rate rises without owned citations may be benefiting from directories, press, or other third-party evidence. A firm with both gains but no inquiries may have a conversion or attribution problem.

Viclaro Atlas supplies the market layer that webmaster tools cannot: repeated buyer-style prompts across multiple assistants, with the named firms and model coverage preserved. Our guide to measuring AI visibility explains why a panel is necessary.

What to do with the new reports this week

Export a baseline before changing content. In Google, record the generative impressions and pages available to your property. In Bing, export cited pages and grounding queries. Label the dates and product scope because both reports are evolving and Bing remains a public preview.

Next, classify the pages by job: service page, practitioner profile, proof page, FAQ, guide, comparison, or location page. Then compare the grounding phrases and visible pages with the buyer situations in your recommendation prompt set. A cited guide that never leads to a recommendation may need a clearer bridge from information to the firm’s relevant capability. A frequently recommended firm with no owned citations may need accessible first-party proof.

Finally, measure again after a defined interval without changing the prompt panel. Webmaster data tells you whether retrieval and exposure changed on Google or Microsoft surfaces. The prompt panel tells you whether assistants changed the shortlist. Analytics and CRM data tell you whether people acted. That is a measurement system; any one dashboard alone is just a window.

What the reporting race means for GEO

The important news is not that two more dashboards exist. It is that AI search is becoming observable at several separate layers. The vocabulary now matters more, not less: impression, citation, mention, recommendation, visit, and conversion describe different events.

Teams that preserve those distinctions can diagnose a real bottleneck. Teams that blend them into a single visibility percentage will produce cleaner charts and weaker decisions. The new first-party reports are valuable inputs. Treat them as the exposure and citation instruments they are, then add independent recommendation monitoring around them.