Microsoft Copilot cited our blog 4,000+ times in 90 days. Here is what it cited, and what it did not send.

Between July and September 2026, Microsoft Copilot cited pages on stellarcast.ai more than 4,000 times. The site was about three months old. On Google, the same pages mostly sat on page five or lower. And the citations sent almost no visitors.

We sell AI visibility measurement, so we measured ourselves. This is what the data shows, what the cited pages have in common, and why "being cited" and "getting traffic" turned out to be two different results for us.

Where the numbers come from

Every figure in this post is first-party data we recorded on 3 October 2026: Bing Webmaster Tools' AI Performance report, which Microsoft introduced as a public preview, for Copilot citations, cited pages and grounding queries; Google Search Console, including its generative AI performance report, for rankings and AI Overviews and AI Mode impressions; and GA4 for visits.

Caveats we want up front, because a case study that hides them is not worth citing:

• Bing labels the report a sample of overall citation activity that it will keep refining "as additional data is processed".
• A citation is one appearance of a page in an answer. It is not a reader, and one answer can be shown to one person or many. We do not know which.
• It covers Microsoft surfaces only (Copilot, Bing's AI summaries and select partners); we call these Copilot citations for short. ChatGPT, Perplexity and Gemini offer no equivalent webmaster report that we know of.
• Some grounding queries are not buyers. One was about "AEO" in the customs sense (Authorized Economic Operator). We left it in the totals; it is a small share.

Curious whether AI engines name you when buyers ask about your category? Get a free visibility audit and see where you stand across ChatGPT, Gemini, Perplexity and Google AI Overviews.

The timeline

Citations started in late July, jumped in the week of 10 August and have held at roughly 500-650 a week since. Bing reported no data for the week of 27 July, which we show as a gap rather than a zero.

What Copilot cited

In our data, citations were extremely concentrated. Two pages earned about 80% of all citations:

Figures are rounded from Bing's page breakdown, which totals slightly less than the weekly series.

Most of our roughly 85 posts were never cited at all.

The questions behind those citations (Bing calls them grounding queries) were mostly buyer questions, not trivia. Where a percentage appears, it is our page's share of the citations Bing shows for those queries.

• Comparison and alternatives questions about one enterprise AI visibility platform, especially its content-automation features.
• "Best tools" questions with a specific requirement: alerting with low false positives (our page had 39% of the citations for that query), real-time misinformation detection (30-42%), detecting agentic traffic from CDN logs (32%).
• Market questions: AI search visibility for companies in Norway (25-33%) and Singapore.

The gap: cited, but not ranked, and not clicked

The same months and the same pages produced three very different results:

Two things stand out.

1. Copilot and Google disagreed. Our second most-cited page, a guide to alternatives for one competitor, ranks deep on Google for that competitor's name, on a results page few searchers are likely to reach. Copilot cited it about 880 times for the same topic. In our data, ranking and being cited behaved like separate results, which suggests the two systems may weigh different inputs.
2. Citations were not traffic. One likely reason: Copilot can answer the question with the page's content, so the reader may have little need to click through. Some Copilot visits may arrive without a referrer and land in "Direct", so "none" is a floor, not an exact count. But nothing in our Direct traffic rose with the citations either.

If our AI visibility reporting had been built only on clicks and referral traffic, it would have shown us almost nothing.

What the cited pages have in common

We cannot prove causation from one site. These are patterns we saw, not a recipe:

• They answer a buying question directly. The two most-cited pages are a "best tools" list and an "alternatives" guide, each opening with a short answer before any detail.
• Facts are specific. Both pages carried prices in a comparison table through the whole period. (The per-vendor check date and engine lists on them today were added on 26 September, after most of these citations.) Specific figures may be easier for an engine to lift into an answer.
• They say when a competitor is the better fit. Each option is framed by who it suits ("best for X" or "if you want X"), including ones where we are not the answer.
• They have a FAQ with matching structured data, but so do all our posts, including the many never cited. In our data, FAQ markup was not what separated them (more on that in does schema markup get you cited by AI).
• Market pages are written for that market. The Norway guide covers Norwegian-language search behavior rather than translating a generic post.
• Nothing blocks the crawlers. Our robots.txt allows all crawlers, including Bingbot, GPTBot, OAI-SearchBot, PerplexityBot and ClaudeBot, and we ping IndexNow after each deploy so Bing can re-crawl quickly.

What did not seem to matter: post count. About 80 posts went up in July and August; citations concentrated on a handful regardless.

How to measure this for your own site

1. Bing Webmaster Tools > AI Performance. Free. Shows Copilot citations, cited pages and grounding queries. Verify your site if you have not.
2. Search Console > Performance > Generative AI (beta). Google's AI Overviews and AI Mode impressions for your pages.
3. GA4 AI Assistant channel for the clicks that do happen. For us it was small. Setup is in how to track ChatGPT and AI traffic in GA4.
4. For ChatGPT, Perplexity and Gemini there is no webmaster report that we know of. The way to know is to ask them your buyers' questions, repeatedly, per market, and record who gets named and which pages get cited. That is what Stellarcast does; the method is in how we measure AI visibility.

We will add our own per-engine numbers for stellarcast.ai on those other engines to this post in an update.

The takeaway

• In our data, being cited and being ranked were different results. Track both.
• Being cited and being clicked were different results too. A citation that names your brand can still shape a buyer's shortlist, even when nobody clicks.
• Citations concentrated. A few pages, each answering a buying question directly, earned nearly all the citations.