Bing Webmaster Tools AI Performance Report: How to Read It

For a business, the value of the report is that the figure comes from the search engine itself, not from an outside tool’s estimate. Microsoft presented it as an early step towards GEO tooling, generative engine optimization, inside Bing Webmaster Tools.

Because Microsoft does not name its partners, the report cannot be used to claim that it includes ChatGPT or any other specific assistant beyond Copilot and Bing. Reading it as “my citations across every assistant” overstates what it counts. The data is also aggregated across all surfaces, so it cannot attribute a citation to Copilot or to a particular partner either.

Bing states that the report only reflects content eligible for its index and that it respects what each site asks for in robots.txt. A page Bing cannot index does not appear, and before reading a zero it is worth checking whether AI crawlers can read the site.

Grounding, in AI answers, is the web material the system uses to support and cite its response, according to Microsoft. Grounding queries are therefore the searches the AI runs, not the person’s: Bing groups them into short phrases, such as “solar energy” or “solar panels” in its own example, and warns that they are not full questions or prompts.

One grounding query can lead to several pages on the site, and one page can appear under several queries. Bing notes that the data is a sample of citation activity, so totals may not match between the pages view, the queries view and the chart.

• Intents classifies each grounding query by intent (informational, commercial, navigational, research, local and others), to show what kind of answer the site is cited in.
• Topics groups related queries into broader themes: in Bing’s example, “solar panels” and “residential solar installation” fall under Solar Energy.
• Citation Share is the percentage of citations the site receives out of all citations shown for the same grounding query. It does not reveal which other sites hold the rest.
• Topics Portfolio gives, for each theme, its approximate size in citations across all sites, how many sites are cited and the queries where other sites are cited and yours is not. It only appears when the site has enough data.

Bing describes Citation Share as an observational metric, not a ranking or a competitive scoreboard, and asks for Topics Portfolio figures to be treated as directional. How to calculate citation share outside Bing, with your own log of questions, is covered separately.

The practical difference is the queries. Google tells you which pages and countries a site appears for in its AI answers; Bing also tells you which phrases its AI used to retrieve it, which is the figure that points to what content to write or expand next.

The 16 pages are the average cited pages metric, which Bing calculates as unique pages cited per day: Microsoft’s AI does not lean on a single Kulturo page but on several at once. The screenshot of the report, with the daily series for citations and cited pages, is in the Kulturo case study.

Citations do not say how many customers arrive, and Kulturo measures that another way: it receives around 7 qualified forms a month from people who say they found it by asking ChatGPT or Claude, according to Kulturo’s data as of 26 October 2026. The two figures are read together, but neither can be derived from the other.

The Bing Webmaster Tools documentation says so in as many words: “It does not represent traffic, clicks, or user engagement.”

The report does not explain why a page is cited either, and it shows no individual answers or exact prompts. What leads an assistant to pick one source over another is covered in how AI models choose their sources.

Beyond Copilot and Bing, where the provider gives sites no report, the technical way to measure is a robot log: a rule in the firewall or CDN that records, without blocking, visits from each provider’s robots, identified by user agent and verified by IP where the provider publishes its ranges. This is what DoGeo sets up for each client, with a count of which pages each robot visits and when.

Each type of robot says something different: the search or indexing robot, which pages enter the assistant’s index; the one that opens a page at a person’s request, that the assistant read it in a conversation; the training robot, only something about training.

Neither Bing’s report nor robot logs record what the assistant writes in its answer. That still takes a fixed set of questions asked every month.