BlogFindings
Microsoft Copilot SEO: Bing names the grounding query, Google names none
Microsoft's AI Performance report in Bing Webmaster Tools, announced on 10 February 2026 and extended on 16 June 2026, shows the grounding queries behind citations in Microsoft Copilot. Google's generative AI performance report, read at its help documentation on 14 August 2026, publishes impressions grouped by page, country, date and device, and no query at all.
This post is a reading of what each vendor published about its own report, not a test of either one. Both Bing announcements were fetched and read on 3 September 2026, and every phrase attributed to Microsoft below is quoted from them. The Google side is taken from the reading of Google's Search Console help documentation published here on 14 August 2026, which has not been repeated since. Lantad has run no analysis of anyone's Bing Webmaster Tools account, holds no sample of Copilot citations, and cannot see either report from outside the dashboards that hold them. What it can add sits earlier in the chain, at the question of whether a crawler could fetch and read the page that either report would otherwise have nothing to count.
In short
- Microsoft Copilot SEO has exactly one first-party data source, the AI Performance report announced for Bing Webmaster Tools on 10 February 2026, which Microsoft says shows how publisher content appears across Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations.
- The metrics Microsoft names in that announcement are Total Citations, Average Cited Pages, Grounding queries, page-level citation activity and visibility trends over time, and not one of them is a click or an impression.
- Microsoft defines Grounding queries as the key phrases the AI used when retrieving content that was referenced in AI-generated answers, and states in the same announcement that the data shown represents a sample of overall citation activity.
- Bing added Citation Share on 16 June 2026, defined by Microsoft as the percentage of citations attributed to your site out of all citations shown across all sites for that same grounding query, and Microsoft states it does not expose competitor domains, represent traffic share, or assign quality scores to content.
- Read out of this repository on 3 September 2026, the crawler registry Lantad ships names 15 robots.txt tokens across 9 vendors and its answer engine registry names 8 engines, and Microsoft appears in neither of them.
| Report | Surfaces it covers | Query shown | Headline metrics | Sampling stated |
|---|---|---|---|---|
| Bing Webmaster Tools AI Performance | Microsoft Copilot, AI-generated summaries in Bing, select partner integrations | Yes, grounding queries | Total Citations, Average Cited Pages, page-level citations | Yes, a sample |
| Search Console generative AI report | AI Overviews and AI Mode in Google Search, and Discover separately | No query dimension | Impressions only, by page, country, date and device | Not stated in the pages read |
| Lantad | 8 answer engine rows, none of them Copilot | Prompts we choose, not user queries | Whether a crawler can fetch, parse and read the page | Not applicable |
What does Microsoft Copilot SEO actually give you to work with?
One report, and it counts citations rather than traffic. Microsoft's announcement of 10 February 2026 says the report shows how publisher content appears across Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations. That sentence is doing more work than it looks. Three distinct surfaces are covered by one number, and the third of them is not enumerated, so a movement in the total cannot be attributed to any one of them from inside the report.
The metrics are named individually and each carries a definition. Total Citations shows the total number of citations that are displayed as sources in AI-generated answers during the selected time frame. Average Cited Pages shows the average number of unique pages from your site that are displayed as sources in AI-generated answers per day over the selected time range. Grounding queries shows the key phrases the AI used when retrieving content that was referenced in AI-generated answers. Page-level citation activity shows citation counts for specific URLs from your site, which Microsoft says makes it easy to see which individual pages are most often referenced. A visibility trend over time completes the set.
Read that list for what is absent rather than what is present. There is no click. There is no impression. There is no position, no click-through rate, and no revenue. Every metric is a count of the same event, which is your URL being displayed as a source inside an answer, sliced by page or by day. That is a narrower thing than a search report and it is also a more honest one: a citation either happened or it did not, whereas an impression inside a generated answer requires somebody to define what counts as being seen.
Microsoft also attaches a caveat that governs everything above it. The announcement states that the data shown represents a sample of overall citation activity. A sample is not a census, and no sampling rate is published, so a Total Citations figure is a lower bound of unknown tightness rather than a count. Anyone building a target or a monthly report on that number is building on a quantity whose relationship to the real one Microsoft has not disclosed. That is not a criticism of shipping it. It is the reason to treat the shape of the trend as the finding and the absolute value as an indication, which is the same discipline our own methodology applies to anything it cannot observe end to end.
-
Total CitationsCount of source displays Citations displayed as sources in AI-generated answers during the selected time frame. -
Average Cited PagesUnique pages per day Average number of unique pages from the site shown as sources per day over the selected range. -
Grounding queriesThe retrieval phrases The key phrases the AI used when retrieving content that was referenced in answers. -
Page-level citation activityCounts by URL Citation counts for specific URLs, so the most referenced pages are visible individually. -
SamplingStated, not quantified The data shown represents a sample of overall citation activity, with no published rate.
Where a Copilot citation comes from before it reaches the report
The word grounding is the useful part of Microsoft's vocabulary here, because it names a step that most reporting hides. A grounding query is not the sentence a person typed into Copilot. It is, in Microsoft's own definition, the key phrase the AI used when retrieving content. Something between the user and the index rewrote the request, and what the report exposes is the rewritten form. That is the phrase your page competed on, which makes it more actionable than the user's wording would have been and also means it is not the user's wording.
Everything upstream of that retrieval is invisible to the report, and it is where a page is usually lost. A URL cannot be retrieved as a source if it was never fetched, never parsed, or fetched and found to hold no readable prose. The report begins at the moment your URL is displayed as a source, so a site with a rendering problem does not appear in it as a warning. It appears as an absence, which reads exactly like a site nobody asked about. Distinguishing those two cases is the entire reason an external check exists, and it is the layer our robots.txt tester and the crawler view of a page operate on.
There is a second gap between the crawl and the citation that no first-party report closes either. Being fetched is not being readable. A page whose text arrives only after client-side hydration, or whose main content is outnumbered by navigation, can be crawled perfectly and still contribute almost nothing an answer engine can quote, which is the property prose parity exists to name. The chain from a request to a citation has at least four links in it, and both first-party reports observe only the last one.
None of this is a criticism of the report for not being a scanner. It is an argument about which question each tool answers. If citations are flat, Bing's report tells you which grounding queries you were cited on and which pages carried them. It cannot tell you whether the pages that were not cited were readable at all. That question is answered by fetching them as a crawler would, which is the half of AI visibility that no logged-in dashboard covers, and the half a publisher can still change on their own site this week.
Flow: Crawler fetches the URL (robots.txt decides) to HTML parsed, text extracted; HTML parsed, text extracted (readable text or none) to Content held for retrieval; Content held for retrieval to Grounding query retrieves it; Grounding query retrieves it to URL displayed as a source; URL displayed as a source (the only visible step) to Counted in the report.
Why no robots.txt line addresses Copilot
A publisher who has just read a Copilot citation report reasonably asks the next question, which is what they can set on their own site to influence it. For every other major vendor the answer starts with a name in a text file. It does not here, and the reason is a design choice Microsoft made years before this report existed.
Read out of this repository on 3 September 2026, the AI crawler registry Lantad ships carries 15 robots.txt product tokens across 9 vendors: three from OpenAI, four from Anthropic, two from Perplexity, and one each from Google, Apple, Meta, ByteDance, Common Crawl and Amazon. Microsoft is not among the nine. That is not an omission in the registry. Microsoft's published guidance for controlling AI use of a page, announced on 22 September 2023, is a meta robots directive carrying NOCACHE or NOARCHIVE rather than a crawler token to disallow, and that announcement does not mention robots.txt at all. The consequences of that choice, including the precedence rule when a page carries both directives, were worked through here in an earlier post on Microsoft's opt out and are not repeated.
What matters for the report is the asymmetry it creates. A Disallow line is one file at a known path, so an external tool can read a site's position on GPTBot or ClaudeBot in a single request, which is what the AI crawler reference is built on. A meta directive lives in the page, so establishing a site's position on Microsoft's AI use requires fetching every page. The control did not get weaker. It got more expensive to survey, and it moved to the layer most site owners edit least, which is a real risk given how often the layer that answers is not the one being edited.
The practical reading is narrow and worth stating plainly. There is no line you can write in robots.txt that Copilot is documented to read, so a robots.txt audit tells you nothing about Microsoft either way, and a clean robots.txt report is not evidence that a site has taken any position at all on Microsoft's AI surfaces. It has taken a position on the nine vendors that publish tokens, which matters for getting cited in ChatGPT and its peers, and Microsoft is a separate question answered in a different file on every page.
- A robots.txt token naming Microsoft's AI use None of the 15 tokens in Lantad's registry on 3 September 2026 belongs to Microsoft, and the 2023 announcement never mentions robots.txt.
- A meta robots directive, NOCACHE or NOARCHIVE Microsoft's documented control, set per page rather than per site, which is why surveying it needs every page fetched.
- A first-party citation report AI Performance in Bing Webmaster Tools, a logged-in dashboard rather than anything an external page fetch can observe.
- A user query, as typed The report publishes grounding queries, which Microsoft defines as the phrases the AI used when retrieving, not what the person asked.
- The other sites cited on the same phrase Microsoft states that Citation Share does not expose competitor domains, so the denominator is visible and its members are not.
What neither first-party report shows, and which tool to use for what
Set the two reports side by side and the disagreement is sharp on one axis. Bing publishes the grounding query. Google, on the reading of its Search Console help documentation published here on 14 August 2026, publishes a single metric, impressions, grouped by pages, countries, dates and devices, with no query dimension in the generative AI performance report at all. One vendor tells a publisher which phrase their page competed on. The other tells them how many times something was shown. Both are first party, and they are not substitutes for one another.
They agree on the thing they both omit. Neither report observes anything before the moment content was retrieved or displayed, so neither can tell a publisher that a page was skipped because a crawler could not read it. That failure produces no row anywhere. It produces the absence of rows, which is indistinguishable in both dashboards from a page nobody wanted. An external fetch is the only way to separate those, which is why the two kinds of tool belong in the same workflow rather than in competition.
Where that leaves this scanner should be said without hedging, because it is inconvenient. Read out of this repository on 3 September 2026, Lantad's answer engine registry holds 8 engines, and Microsoft Copilot is not one of them, so nothing Lantad runs observes a Copilot answer. Anyone who needs Copilot citation counts should use Bing Webmaster Tools, which is first party and gives a number no external tool can derive. Anyone who needs Copilot answers monitored on a schedule needs a product that puts prompts to models, and of the ten AI visibility products read at their own pages on 17 August 2026, six do exactly that. Lantad's own coverage of the engines it does ask was set out when it published that 3 of its 8 engines cannot search the web.
The honest summary is that Microsoft Copilot has better first-party citation reporting than Google does on the one dimension publishers ask for most, and worse site-level controls, and that neither fact is visible from outside the login. What is visible from outside is whether the page an answer engine would need is fetchable, parseable and mostly content rather than navigation, and that remains true across AI Overviews, Copilot and every engine in between, because all of them read the page before they can quote it. Measuring what answer engines actually say and measuring whether they can read you are two jobs, and a report that does one of them well is not evidence about the other.
-
Which phrase was I cited on?Bing only Grounding queries are published by Bing's report. Google's generative AI report carries no query dimension. -
How many times was I shown?Both, differently Bing counts citations as a sample. Google counts impressions. The two are not the same event and cannot be added. -
What share of the phrase was mine?Bing only Citation Share, added 16 June 2026, with competitor domains withheld by design. -
Could a crawler read the page at all?Neither report Both begin at retrieval or display, so an unreadable page shows up as absence rather than as a fault. -
What does Copilot say about my brand?Not this scanner Lantad's engine registry held 8 engines on 3 September 2026 and Copilot is not one of them.
Lantad
Published .
Anyone working on Microsoft Copilot SEO meets the problem that stalled every other AI surface for two years. The answers are assembled somewhere you cannot observe, and the vendor is the only party holding the numbers. Microsoft began handing some of those numbers back on 10 February 2026, when AI Performance arrived in Bing Webmaster Tools as a public preview, and extended it on 16 June 2026. Google announced its own version for AI Overviews and AI Mode on 3 June 2026. Those two reports are the only first-party views of AI citation a site owner can log in and read, and they do not report the same things.
Common questions
Is there a Microsoft Copilot SEO report?
Yes, one. AI Performance in Bing Webmaster Tools, announced on 10 February 2026, which Microsoft says shows how publisher content appears across Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations. It reports citations rather than traffic, and the metrics named in that announcement are Total Citations, Average Cited Pages, Grounding queries, page-level citation activity and visibility trends over time.
What is a grounding query?
Microsoft defines it as the key phrases the AI used when retrieving content that was referenced in AI-generated answers. That is not the sentence a person typed into Copilot: the request was rewritten before retrieval, and the report exposes the rewritten form. It is the phrase your page competed on, which is more useful for content work and is also not the user's own wording.
Can I block Copilot in robots.txt?
There is no documented Microsoft AI token to write a Disallow line against. Microsoft's guidance, published on 22 September 2023, is a meta robots directive carrying NOCACHE or NOARCHIVE, set in the page rather than in one file at a known path. Read on 3 September 2026, none of the 15 robots.txt tokens in Lantad's crawler registry belongs to Microsoft, and the nine vendors that do publish tokens are OpenAI, Anthropic, Perplexity, Google, Apple, Meta, ByteDance, Common Crawl and Amazon.
Does Lantad track Microsoft Copilot?
No. Read out of this repository on 3 September 2026, Lantad's answer engine registry holds 8 engines and Copilot is not one of them, so nothing Lantad runs observes a Copilot answer. For Copilot citation counts the right tool is Bing Webmaster Tools, which is first party. What Lantad measures instead is whether a crawler can fetch, parse and read the page that any of these reports would otherwise have nothing to count.
See what AI can read on your site
Run a free scan and get a graded report of exactly what AI crawlers can and cannot read, with ranked fixes.
AI visibility
AI visibility is whether AI systems can access, read, and faithfully represent your web pages, the prerequisite for being cited in an AI answer.
AI crawler
An AI crawler is an automated bot, such as GPTBot, ClaudeBot, or PerplexityBot, that fetches web pages so an AI system can read, train on, or cite what those pages contain.
Microsoft's AI opt out is a meta tag, not a robots.txt token
Every other major vendor gave site owners a robots.txt name to disallow. Microsoft's published guidance for keeping a page out of its AI answers is NOCACHE or NOARCHIVE in a meta robots tag, which is why none of the 15 tokens in our crawler registry belongs to Microsoft.
Search Console's generative AI report counts impressions and nothing else
Google's generative AI performance report, read at its Search Console help documentation on 14 August 2026, offers one metric and four dimensions. There is no clicks column, no click-through rate, no average position and no query dimension, and the Search and Discover versions count an impression by opposite rules.
AI visibility tracking: 3 of the 8 engines we ask cannot search the web
Read out of this repository on 29 August 2026, the answer engine registry Lantad ships carries 8 engines. Five are recorded as performing live retrieval and returning the sources they grounded in. Three are recorded as having no web access and no sources at all, and one of those three is the only engine a free run asks.
