Bing Webmaster Tools AI Performance report review
first party, free, and structurally underrated
By Jānis Plūme, Founder, Outbound Pros · 9 min read · 2026-08-06
Quick answer
The AI Performance report in Bing Webmaster Tools is the only place in this review where the numbers are measured rather than inferred, because Microsoft is reporting on its own surfaces from its own logs, and it is free. It covers Microsoft Copilot, AI generated summaries in Bing and selected partner integrations, and it exposes one field that exists nowhere else in the stack: grounding queries, the phrases the AI actually used when it went looking for your content. It cannot tell you anything about ChatGPT app answers, Claude, Perplexity or Gemini, which is why it is a foundation rather than a complete answer.
What the report is and who builds it
It is built by Microsoft and it lives inside Bing Webmaster Tools, which is free and requires domain verification to use. Microsoft shipped the AI Performance report in public preview in February 2026. It covers Microsoft Copilot, AI generated summaries inside Bing, and selected partner integrations, and it reports total citations, average cited pages, page level citation activity, and grounding queries.
Grounding queries are the reason this report belongs at the top of the list rather than at the bottom. Microsoft defines them as the phrases the AI actually used when retrieving your content. Classic search analytics tell you what a person typed. Grounding queries tell you what the model reformulated that into before it went looking, which is a different string and usually a more specific one. If you have wondered what Google means by query fan out, its own term for the set of concurrent related queries a model generates to fetch additional results, this is the closest anyone outside a platform can currently get to watching it happen against their own pages.
The strategic reason to care about Microsoft here has nothing to do with the size of Bing as a consumer destination, which is the mental shortcut that causes teams to skip it. Bing is the index that feeds ChatGPT candidate retrieval and Copilot answers. Being absent from it is not a small regional gap, it is a gap in the retrieval layer under one of the surfaces you most want to appear in.
Who this genuinely suits
Everyone, without qualification, which is not a sentence we write about any other product in this review. It is free, it takes under an hour including domain verification, and it produces measured first party data. There is no company for which the correct decision is to skip it.
- Any team about to start AI search work, because a baseline established in week one is worth more than a month of tactics
- Anyone being pitched a paid AI visibility platform, since this is the free control against which to read the vendor demo
- Content teams who want to know which specific pages are being cited rather than whether the brand is doing well in general
- Anyone writing content briefs, because grounding queries are the closest thing to keyword data that exists on the AI side
- Sites where IndexNow is not yet wired up, since this is the same console and the integration removes discovery delay on the index that feeds ChatGPT search
Microsoft has also published guidance on earning citations that is short, unglamorous and consistent with every independent measurement we have seen: improve structure and clarity with headings, tables and question and answer sections, support claims with evidence and data, keep content current, and keep the representation consistent across text, images and video. That is the whole of gate two stated by a platform rather than by a vendor selling you a method, and it costs nothing to read.
Where it is weak or the wrong choice
Coverage is the obvious limit and it is a hard one. This report says nothing about answers generated inside the ChatGPT application, nothing about Claude, nothing about Perplexity and nothing about Gemini. For those surfaces no first party reporting exists at all, from anyone, and the only honest method is a frozen panel of twenty to thirty prompts run identically every month in fresh sessions. Anyone selling you a measured ChatGPT visibility figure is selling you a sample with a confident label on it.
It also will not tell you how you were described, which is often the more urgent question. Citation counts go up while the accompanying description says you serve a market you exited two years ago. That is an entity problem, it is invisible in a citation report, and it is the failure mode that costs the most before anyone notices it.
It is preview stage reporting, which means the fields, the definitions and the interface can change, and comparing a figure from six months ago against today carries a risk the interface will not warn you about. Screenshot your baseline and record what each field meant on the day you recorded it. That habit costs nothing and it is what separates a measurement programme from a set of numbers you no longer trust.
And, like everything else in this review, it sees nothing about whether your pages return real HTML to a non rendering crawler. A page that is indexed by Bing and citable by Copilot can still be an empty container to the search bots that go and fetch a URL to extract a quotable sentence, because Bingbot renders and those agents do not. Being indexed and being quotable are separate problems, and passing the first tells you nothing about the second.
| Dimension | Rating | What that means in practice |
|---|---|---|
| Data provenance | Best in class | Measured, first party, from the platform generating the answers. Nothing else in this review can make that claim, and nothing else ever will for a surface it does not own. |
| Engine coverage | Limited by design | Microsoft Copilot, AI summaries in Bing and selected partner integrations. Silent on ChatGPT app answers, Claude, Perplexity and Gemini, and that silence is honest rather than a gap it pretends to fill. |
| Crawler access and rendering, gate one | Not covered | Bingbot behaviour is not AI crawler behaviour. Being indexed here says nothing about whether a non rendering agent received your body copy. |
| Source and corroboration discovery, gate three | Not covered | Reports on your own URLs only. For the map of third party sources being cited in your category, you need a tracker built for it. |
| Method transparency | Strong | Field definitions come from the platform that generates the data, including a stated definition of grounding queries. No modelling layer between you and the number. |
| Operating burden | Very light | Under an hour to set up including domain verification. The only real cost is remembering to open it, which is where most of these programmes die. |
Disclosure: we are not a neutral party
InboundPros is run by the same person who runs the outbound agency in this group, and the group sells managed outbound rather than inbound retainers. That is the conflict, and on this particular page it points in an unusual direction: we are telling you that the most valuable instrument in the category is free, from Microsoft, and takes an hour. There is no affiliate revenue anywhere on this site, and none of the paid tools reviewed here pays us. The reason to trust this entry more than the others is that it is the one where we had the least to gain from any answer. The reason to read the paid reviews carefully anyway is the same disclosure applied in reverse. If your actual constraint turns out to be pipeline rather than visibility, the group publishes a free GTM audit and that is the one place on this page where we are selling you something.
Questions we get asked about the Bing AI report
Is it worth setting up if my audience does not use Bing?
Yes, and the framing is the problem rather than the answer. Bing is not only a consumer search destination in this context, it is the index that feeds ChatGPT candidate retrieval and Copilot answers. Ranking in Bing appears close to necessary for those surfaces and is clearly not sufficient. You are not setting it up for Bing traffic. You are setting it up because it is the only window onto the retrieval layer that sits underneath surfaces you care about a great deal.
What is a grounding query and how do I use one?
Microsoft defines grounding queries as the phrases the AI actually used when retrieving your content, which is not the same string a person typed. Use them the way you would use keyword data, with better provenance: they tell you the exact form of the question your page was retrieved for, which makes them the strongest available input to a content brief. If a grounding query is more specific than anything you have written a section about, that is a section to write.
Does this replace a paid AI visibility tool?
It replaces the Microsoft rows of one and nothing else. Everything a paid tool shows you about ChatGPT, Claude and Perplexity is prompt sampling, which is a legitimate method and a different kind of object from this report. The sensible sequence is to switch this on, switch on the Search Console generative AI performance report, run a frozen prompt panel by hand for a quarter, and only then decide whether a paid tool is removing effort or adding reassurance.
How quickly will a new page show up?
Unknown in general, and nobody outside the platforms can tell you honestly. What is controllable is discovery delay: submit sitemaps in the same console and wire up IndexNow on publish and on update. Microsoft recommends it directly, it is a one time integration, and it covers both Copilot and ChatGPT candidate retrieval. Beyond that, be careful with anyone quoting you a pickup time, because there is no published crawl frequency for the AI search agents and inferring one from a handful of observations is not a measurement.
Switch on the free first party reports this week. Then spend ten minutes confirming that a non rendering crawler gets your actual copy, because every number above assumes it does.
Last updated: 2026-08-06
See what a crawler sees
on your own site
Paste a URL and get the extractability read: what an assistant can actually retrieve, and what it cannot.
Free. No signup, no email capture.