Which Pages Does Microsoft Copilot Cite? A Study of 328,700 AI Citations to One Website
Microsoft Copilot and its partners cited my website 328,700 times in three months, according to Bing Webmaster Tools, rising from 539 citations a day in late July to 11,225 a day in early October. Most of them didn't go to pages about my services. They went to plain answers to practical questions. About one in seven of the top searches wasn't in English, and one page took 30.2% of the total. Searches I wrote a matching section for gained share in 13 of 15 cases, but searches I left alone rose too, so the gain from the sections is smaller than it first looks.
Published 11 October 2026 by Lilach Bullock. Data from Bing Webmaster Tools, covering 11 July to 8 October 2026, read on 11 October 2026.
Everyone wants to know how to get cited by AI. Most advice on it is guesswork, because the AI companies show you very little. Microsoft is the exception. Bing Webmaster Tools has an AI Performance report that shows how often Copilot and its partners cite your pages, which pages they cite, and the searches Copilot ran before it cited you.
So I opened mine and went through three months of it: 328,700 citations, 1,319 cited pages and 1,468 searches. Here's what Copilot cites, what it's looking for when it does, and what changed the numbers.
Key statistics
- Citations grew 21 times in under three months. From an average of 539 a day between 11 and 31 July to 11,225 a day between 1 and 8 October. The busiest day was 30 September, with 13,500.
- One page took 30.2% of all citations. A guide to games you can play without the internet earned 99.3K of the 328.7K citations. The next 11 most cited pages took 12.6% between them.
- 23 of the top 154 searches were in a language other than English, across 9 languages, and they earned 13.4% of the citations in that list.
- Matching the query helps, but less than it first looks. Of the 15 searches I wrote a matching section for on 2 September, 13 gained share by the 11 October read (median change +5.2 points), and the best moved from 11.7% to 31.5%. But 7 of the 13 comparable searches I left alone also gained (median +4 points).
All 9 findings
- Citations grew 21 times in under three months. From an average of 539 a day between 11 and 31 July to 11,225 a day between 1 and 8 October. The busiest day was 30 September, with 13,500.
- The number of pages being cited grew 11 times. From 60 pages a day in late July to 653 in early October (a high of 742 on 7 October). Across the three months Copilot cited 1,319 different pages for 1,468 different searches.
- One day in September changed the curve, and I can't prove why. Citations went from 3,900 on 15 September to 8,400 on 16 September and 13,300 on 17 September, but the number of pages cited barely moved (458, 462, 428). The same pages were simply being cited far more often.
- One page took 30.2% of all citations. A guide to games you can play without the internet earned 99.3K of the 328.7K citations. The next 11 most cited pages took 12.6% between them.
- Copilot searches like a person typing into a search box, not like a person asking a question. The median search it ran was 4 words long, and only 13 of the top 154 searches (8.4%) were worded as a question.
- It looks for evaluations. 23 of the top 154 searches (14.9%) asked Copilot to judge something, with words like evaluation, review, legit, best or compare, as in "Wix marketing strategy evaluation e-commerce".
- 23 of the top 154 searches were in a language other than English, across 9 languages, and they earned 13.4% of the citations in that list.
- Practical how-to answers beat marketing pages. Games, social media how-tos, remote work and product review sites made up 90.6% of citations in the top 154 searches.
- Matching the query helps, but less than it first looks. Of the 15 searches I wrote a matching section for on 2 September, 13 gained share by the 11 October read (median change +5.2 points), and the best moved from 11.7% to 31.5%. But 7 of the 13 comparable searches I left alone also gained (median +4 points).
How Copilot citations grew from July to October
These are daily averages for each period. A citation is one time Copilot or one of its partners linked to one of my pages as a source in an answer.
| Period | Days | Citations a day | Chart | Pages cited a day | Citations per cited page |
|---|---|---|---|---|---|
| 11 to 31 July | 21 | 539 | 60 | 9.0 | |
| August | 31 | 1,141 | 130 | 8.8 | |
| 1 to 15 September | 15 | 2,420 | 315 | 7.7 | |
| 16 to 30 September | 15 | 10,420 | 523 | 19.9 | |
| 1 to 8 October | 8 | 11,225 | 653 | 17.2 |
Two things stand out. Growth wasn't a straight line: between 27 July and 6 August citations fell to an average of 282 a day before recovering. And the jump in the second half of September came from citations per page, which went from 7.7 to 19.9, rather than from more pages being cited.
The 16 September step change
| Date | Citations | Pages cited |
|---|---|---|
| 14 September | 3,500 | 420 |
| 15 September | 3,900 | 458 |
| 16 September | 8,400 | 462 |
| 17 September | 13,300 | 428 |
| 18 September | 11,400 | 455 |
I've looked for the cause and I can't prove it. I haven't found a change on my side that explains it. The timing lines up with the offline games pages growing, but lining up isn't proof, and Microsoft doesn't publish changes to how Copilot picks its sources. What the data does show is that the same few hundred pages started being cited more than twice as often, almost overnight, and the higher level held for the rest of the period.
Which pages get cited
These were the 12 most cited pages over the three months, as a share of all 328.7K citations.
| Page | Citations | Share of all citations | Chart |
|---|---|---|---|
| Games you can play without the internet | 99.3K | 30.2% | |
| How to find archived Facebook posts | 5.7K | 1.7% | |
| Best product review sites before you buy | 5.4K | 1.6% | |
| How to create Twitter Moments | 4.2K | 1.3% | |
| Instagram font generator | 3.5K | 1.1% | |
| Wix marketing strategy | 3.5K | 1.1% | |
| Where to find a Facebook Page ID | 3.5K | 1.1% | |
| Free content creation tools | 3.3K | 1% | |
| Best product review sites for shoppers | 3.2K | 1% | |
| Free offline games to download | 3.2K | 1% | |
| Cutting the cost of a marketing tech stack | 3.1K | 0.9% | |
| Squarespace marketing strategy | 2.8K | 0.9% |
That top row is the biggest risk in the whole report. One page earning 30.2% of citations means a single change to that page, or to how Copilot treats that kind of question, could take away almost a third of the total. The other 1,318 cited pages share the rest, and that spread is what I'd want on any site.
What kinds of questions get cited
Bing shows 1,468 searches that Copilot ran before citing my pages. I sorted the 154 with the most citations into five groups. Between them they account for 126.7K citations.
| Kind of search | Searches | Citations | Share | Chart | Examples |
|---|---|---|---|---|---|
| Offline and no-wifi games | 72 | 86.2K | 68% | "offline games free"; "no wifi games"; "free games no wifi needed" | |
| Social media how-tos | 37 | 21.3K | 16.8% | "find id facebook"; "highlighting brand milestones in pinned top tweets"; "instagram guest mode" | |
| Marketing, AI tools and platform evaluations | 26 | 11.9K | 9.4% | "Wix marketing strategy evaluation e-commerce"; "signs of an ineffective AI strategy"; "affordable alternatives to Meta Ads for small businesses" | |
| Remote work, jobs and pay | 13 | 4.8K | 3.8% | "Amazon remote work from home jobs evaluation"; "are online tutor jobs legit uk"; "remote data entry pay uk" | |
| Product review sites | 6 | 2.6K | 2.1% | "product reviews sites"; "best trusted buying guide websites"; "most reliable sources for product roundups" |
Games dominate because of one cluster of pages, but look at the rest. Copilot cites practical answers: how to find a Facebook Page ID, whether an online tutoring job is legit, how to pin a tweet that shows off a milestone, which review sites to trust before you buy. It also cites evaluations of business tools, with searches like "Squarespace marketing strategy evaluation" and "affordable alternatives to Meta Ads for small businesses". These are the searches of someone about to make a decision.
How Copilot words its searches
| Measure | Top searches | Share of those searches | Share of their citations |
|---|---|---|---|
| Worded as a question (how, what, are, is and so on) | 13 | 8.4% | 3.1% |
| Asking for a judgement (evaluation, review, legit, best, compare, vs, reliable, trusted) | 23 | 14.9% | 8.8% |
The median search was 4 words. Many read like something a machine wrote, not a person: "Amazon Jobs online recruitment platform review remote jobs" or "mine emails support tickets forums buyer language no keyword volume". That's because these are the searches Copilot runs to find sources, after it has turned the user's question into keywords. The page that wins is the one whose heading and first sentence match those keywords.
Copilot cites in other languages
23 of the top 154 searches weren't in English. Most of my site is written in English, yet these searches earned 13.4% of the citations in the list.
| Language | Searches in the top 154 | Citations | Example search |
|---|---|---|---|
| Spanish | 2 | 9,126 | juegos sin internet gratis |
| German | 4 | 2,290 | Marketing-Tech-Stack Kostenreduzierung Strategien |
| French | 7 | 1,695 | jeu sans wifi |
| Russian | 2 | 1,134 | игры без интернета |
| Arabic | 2 | 966 | العاب بدون نت |
| Italian | 2 | 644 | giochi senza internet |
| Portuguese | 2 | 469 | jogo sem internet |
| Dutch | 1 | 354 | instagram lettertype |
| Turkish | 1 | 273 | internetsiz oyunlar |
My site has a handful of pages written in Spanish, French, Italian, Portuguese and Turkish, and one English guide carries a short summary in German. Bing doesn't show which page each search led to, so I can't match every search to a page. If your customers search in more than one language, a few lines in their language near the top of the right page can be enough to be cited.
What happened when I wrote sections to match Copilot's searches
On 2 September I picked searches where Copilot was already citing my pages and I wanted a bigger share. For each one I added a short section to the page it was already citing: a heading worded like the search and 120 to 200 words answering it. Nothing was removed. Bing shows a share figure next to every search, and this table compares it across three reads.
| Search | Section added to | Share on 2 Sep | Share on 15 Sep | Share on 11 Oct | Change (points) | Note |
|---|---|---|---|---|---|---|
| highlighting brand milestones in pinned top tweets | pinned tweets: brand milestones (plus 4 internal links) | 11.7% | 21.1% | 31.5% | +19.7 | |
| are online tutor jobs legit uk | online tutoring jobs | 23.4% | 29.5% | 40% | +16.6 | |
| how to find top trending hashtags on facebook reels | trending Facebook Reels hashtags | 18.6% | not in the top 25 | 31.1% | +12.5 | |
| are online tutoring jobs legitimate and how much pay | online tutoring jobs | 19% | 23.3% | 30.5% | +11.5 | |
| Squarespace product marketing promotion evaluation | Squarespace marketing strategy | 11.3% | 14.3% | 20.1% | +8.8 | |
| Marketing Tech Stack Kosten reduzieren Strategien | German summary on the tech stack guide | 24.3% | 24.3% | 32.1% | +7.8 | |
| top ways to grow a facebook group organically | private Facebook groups guide | 13.4% | 14.8% | 19.2% | +5.8 | |
| Amazon Jobs online recruitment platform review remote jobs | Amazon remote jobs | 43% | 43.5% | 48.2% | +5.2 | new post topic also queued 2 Sep |
| signs of an ineffective AI strategy | why AI projects fail | 14.8% | not in the top 25 | 18.4% | +3.6 | new post topic also queued 2 Sep |
| Amazon Jobs platform review online recruitment job search | Amazon remote jobs | 15.2% | 17.7% | 17.6% | +2.4 | new post topic also queued 2 Sep |
| product reviews sites | best product review sites | 19.7% | 15.9% | 20.7% | +1 | |
| freelance writing jobs uk for beginners | freelance writing jobs UK | 16.7% | 17% | 17.2% | +0.5 | |
| statsocial | StatSocial review | 14.7% | not in the top 25 | 15% | +0.3 | |
| content creation tools | free content creation tools | 10.2% | not in the top 25 | 8.9% | -1.3 | |
| how to use Twitter Moments (query reworded)* | Twitter Moments guide | 15.7% | 13.1% | 11.2% | -4.5 |
* I've reworded that search here; the real query uses a word I avoid on this site. It's the one that fell, and it tells me a matching section isn't enough when the page underneath it doesn't answer the question well.
For comparison, these are searches from the same 2 September list that got no new section. Several gained too, which is why I don't claim the sections caused every point. The sectioned searches moved a median of +5.2 points. The 13 searches with no section and no other change I know of moved +4; all 19 together moved +4.3. Some sit on a page that got a section for a different search, and three were topics I queued for new posts the same day, so I've marked them.
| Search (no new section) | Share on 2 Sep | Share on 11 Oct | Change (points) | Note |
|---|---|---|---|---|
| create your own twitter moment | 23.6% | 39.3% | +15.6 | same page as a sectioned search |
| showcasing user stories in pinned tweets | 38.5% | 51.2% | +12.7 | |
| soliciting brand feedback with pinned twitter forms | 21.8% | 34.2% | +12.4 | new post topic queued 2 Sep |
| offline games free no wifi | 25.9% | 37.9% | +11.9 | |
| Wix marketing strategy evaluation e-commerce software | 33.2% | 43.1% | +9.9 | |
| Wix marketing strategy evaluation e-commerce | 33.6% | 42.5% | +8.9 | |
| creating twitter moments | 16.1% | 23.5% | +7.4 | same page as a sectioned search |
| pinning brand disclaimers on twitter profiles | 16.2% | 22.9% | +6.7 | new post topic queued 2 Sep |
| compare followers and following instagram | 18.2% | 23.1% | +4.9 | |
| Amazon remote work from home jobs evaluation | 33.3% | 37.7% | +4.3 | |
| highlighting loyal fans in pinned twitter showcases | 24.4% | 28.5% | +4.1 | new post topic queued 2 Sep |
| threads icon | 24% | 28% | +4 | |
| Squarespace marketing strategy evaluation | 32.4% | 30.9% | -1.5 | |
| tips for writing compelling video captions | 24.3% | 21.4% | -3 | |
| AI churn prevention workflows enterprise customer success | 35.6% | 32.3% | -3.3 | |
| top twitter moment creation tips | 23.6% | 20.2% | -3.4 | same page as a sectioned search |
| FlexJobs high paying jobs review | 39.7% | 35.9% | -3.8 | |
| compare instagram followers and following | 28.5% | 22.7% | -5.8 | |
| remote data entry pay uk | 50.2% | 43.7% | -6.5 |
I left out 5 searches whose citation count was identical in both reads, because no new citations means no evidence either way.
What this means for your business
I've been in marketing for 21 years, and this is the first time a search company has shown me, page by page, what its AI chooses to quote. Here's what I'd do with it.
- Open Bing Webmaster Tools and look at the AI Performance report. It's free, and it lists the searches Copilot ran before it cited you. That list is a better brief than any keyword tool, because it's what the AI is looking for.
- Write headings that match the search, not your brand voice. Copilot searches in keywords. Put those words in a heading and answer the question in the first sentence under it.
- Add to the pages that are already being cited. Every section in my test went on a page Copilot already trusted. That's faster than starting a new page.
- Write honest evaluations of the tools your customers are choosing between. Copilot keeps asking for evaluations and reviews. A page that judges something fairly, with reasons, is what it looks for.
- Answer in your customers' languages. A short summary in Spanish, French or Arabic can win citations an English page never would.
- Watch your concentration. If one page earns a third of your citations, you have one point of failure. Spread the answers across more pages.
- Check every month. The share figures move, and a section that worked in September can slip by October.
If you want help with this, my get cited by AI work covers exactly this, and my guide to getting cited by Microsoft Copilot has the step by step version. For Google's side of the same question, see my AI Overviews study of 400 local service searches, and every study I publish is listed on my research page.
Methodology
- Source: the AI Performance report in Bing Webmaster Tools for one website (lilachbullock.com). Citation sources are listed by Bing as Microsoft Copilots and partners.
- Window: the report's three-month view, 11 July to 8 October 2026, read on 11 October 2026. Earlier reads on 2 September (window 2 June to 31 August) and 15 September (window to 13 September) are used for the before and after comparison.
- Totals: 328.7K citations, an average of 257 cited pages a day, 1,319 cited pages and 1,468 searches. Bing rounds large daily figures (for example 13.3K), so the daily rows add up to 329,090, within 1% of the total.
- Searches: the 154 searches with the most citations, out of 1,468. I sorted them into five groups by topic and by language by hand rules that are written into the build of this page.
- Share: the share percentage Bing shows next to each search. Each read covers a rolling three-month window, so the 2 September and 11 October windows overlap from 11 July to 31 August. Compare the direction of the change, not the exact size.
- Sampling: Bing says the report shows a sample of overall activity and that figures may be refined as more data is processed.
- Limits: this is one website. The sections in the before and after test weren't a controlled experiment, and other things changed on the site at the same time. Treat the findings as a detailed picture of how Copilot treated one site, not a rule for every site.
Daily data
Show all 90 days (citations and pages cited, as shown by Bing Webmaster Tools)
| Date | Citations | Pages cited |
|---|---|---|
| 2026-07-11 | 491 | 56 |
| 2026-07-12 | 449 | 50 |
| 2026-07-13 | 521 | 68 |
| 2026-07-14 | 540 | 55 |
| 2026-07-15 | 490 | 64 |
| 2026-07-16 | 520 | 66 |
| 2026-07-17 | 621 | 66 |
| 2026-07-18 | 595 | 70 |
| 2026-07-19 | 595 | 70 |
| 2026-07-20 | 731 | 81 |
| 2026-07-21 | 837 | 84 |
| 2026-07-22 | 620 | 73 |
| 2026-07-23 | 742 | 80 |
| 2026-07-24 | 890 | 85 |
| 2026-07-25 | 638 | 72 |
| 2026-07-26 | 663 | 75 |
| 2026-07-27 | 332 | 36 |
| 2026-07-28 | 166 | 28 |
| 2026-07-29 | 283 | 29 |
| 2026-07-30 | 326 | 32 |
| 2026-07-31 | 277 | 22 |
| 2026-08-01 | 190 | 18 |
| 2026-08-02 | 419 | 24 |
| 2026-08-03 | 251 | 27 |
| 2026-08-04 | 247 | 31 |
| 2026-08-05 | 301 | 35 |
| 2026-08-06 | 313 | 28 |
| 2026-08-07 | 816 | 78 |
| 2026-08-08 | 736 | 69 |
| 2026-08-09 | 838 | 68 |
| 2026-08-10 | 1,000 | 92 |
| 2026-08-11 | 681 | 65 |
| 2026-08-12 | 1,000 | 112 |
| 2026-08-13 | 1,200 | 105 |
| 2026-08-14 | 1,000 | 104 |
| 2026-08-15 | 994 | 89 |
| 2026-08-16 | 1,200 | 118 |
| 2026-08-17 | 1,500 | 145 |
| 2026-08-18 | 1,500 | 154 |
| 2026-08-19 | 1,500 | 138 |
| 2026-08-20 | 1,600 | 160 |
| 2026-08-21 | 751 | 88 |
| 2026-08-22 | 723 | 78 |
| 2026-08-23 | 603 | 74 |
| 2026-08-24 | 1,200 | 165 |
| 2026-08-25 | 2,300 | 254 |
| 2026-08-26 | 2,100 | 285 |
| 2026-08-27 | 2,500 | 292 |
| 2026-08-28 | 2,200 | 296 |
| 2026-08-29 | 1,700 | 266 |
| 2026-08-30 | 1,800 | 271 |
| 2026-08-31 | 2,200 | 310 |
| 2026-09-01 | 2,000 | 265 |
| 2026-09-02 | 2,300 | 301 |
| 2026-09-03 | 2,300 | 275 |
| 2026-09-04 | 2,200 | 261 |
| 2026-09-05 | 1,600 | 232 |
| 2026-09-06 | 1,700 | 245 |
| 2026-09-07 | 1,800 | 269 |
| 2026-09-08 | 2,100 | 304 |
| 2026-09-09 | 2,700 | 348 |
| 2026-09-10 | 2,800 | 355 |
| 2026-09-11 | 2,800 | 349 |
| 2026-09-12 | 2,000 | 306 |
| 2026-09-13 | 2,600 | 340 |
| 2026-09-14 | 3,500 | 420 |
| 2026-09-15 | 3,900 | 458 |
| 2026-09-16 | 8,400 | 462 |
| 2026-09-17 | 13,300 | 428 |
| 2026-09-18 | 11,400 | 455 |
| 2026-09-19 | 9,600 | 382 |
| 2026-09-20 | 9,400 | 431 |
| 2026-09-21 | 10,100 | 500 |
| 2026-09-22 | 9,900 | 511 |
| 2026-09-23 | 9,800 | 565 |
| 2026-09-24 | 9,800 | 581 |
| 2026-09-25 | 9,300 | 557 |
| 2026-09-26 | 8,500 | 510 |
| 2026-09-27 | 9,300 | 514 |
| 2026-09-28 | 12,200 | 605 |
| 2026-09-29 | 11,800 | 664 |
| 2026-09-30 | 13,500 | 682 |
| 2026-10-01 | 13,100 | 663 |
| 2026-10-02 | 11,900 | 632 |
| 2026-10-03 | 9,100 | 564 |
| 2026-10-04 | 8,400 | 542 |
| 2026-10-05 | 11,200 | 699 |
| 2026-10-06 | 13,000 | 663 |
| 2026-10-07 | 12,500 | 742 |
| 2026-10-08 | 10,600 | 716 |
Cite this study
Lilach Bullock (2026). Microsoft Copilot Citations Study 2026: what 328,700 AI citations to one website show. https://www.lilachbullock.com/microsoft-copilot-citations-study/
You're welcome to quote any figure on this page in articles, reports and presentations. Please name the study and link to the page above so readers can see the method.
Frequently asked questions
Which pages does Microsoft Copilot cite?
In this study, Copilot cited pages that answer a practical question directly: how-to guides, honest evaluations of tools and platforms, and lists people use before a decision. Across three months it cited 1,319 different pages on one website, and one page earned 30.2% of all citations.
How can I see if Copilot cites my website?
Bing Webmaster Tools has a free AI Performance report. It shows how many times Copilot and its partners cited your pages, which pages they cited and the searches Copilot ran before citing them.
Does Copilot search in full questions?
Mostly not. In this study the median search Copilot ran was 4 words long and only 8.4% of the top searches were worded as a question. So write headings that use those keywords and answer the question in the first sentence.
Can adding a section to an existing page increase AI citations?
Often, yes. Of 15 searches that got a matching section on 2 September, 13 gained share by 11 October, with a median change of +5.2 points. Comparable searches without a new section also moved, by a median of +4 points, so this isn't proof of cause.
Does Copilot cite pages in other languages?
Yes. 23 of the top 154 searches in this study were in languages other than English, across 9 languages including Spanish, French, Russian, Arabic and German.
Can I use these statistics?
Yes. Please credit Lilach Bullock and link to https://www.lilachbullock.com/microsoft-copilot-citations-study/ so your readers can see the method and the full tables.