Copilot Citation Factors, Graded by Evidence

Copilot answers draw on Bing's index. Microsoft documents Bing as the backing store for Copilot, for AI summaries in Bing, and for its developer product Grounding with Bing Search. That gives a site owner one clear dependency: if Bing doesn't hold your page, Copilot has nothing of yours to retrieve.

It helps to split the path into stages:

• Access: can bingbot fetch the page, and does Bing keep it in the index? You control most of this.
• Retrieval: does the page enter the answer's context for a given question? Microsoft controls this.
• Generation: which retrieved pages get cited? The model decides.

The retrieval stage is hard to observe. AI engines rewrite a user's prompt into their own search queries. Bing Webmaster Tools is the only place in the industry that shows those queries, in its AI Performance report, which went into public preview on 10 February 2026 and was expanded in June 2026. Microsoft notes the grounding query data is a sample. For what tracking tools can and can't see, read whether AI visibility tools can track Copilot answers.

Grades reflect the strength of evidence, not how easy the factor is to act on.

The pattern matches other engines. The levers with evidence remove blocks. Nothing additive has credible support. The same grading applied to OpenAI is in ChatGPT visibility factors, graded by evidence.

Microsoft documents that Bing's index backs Copilot. It does not publish how strongly index presence, or Bing ranking, predicts a Copilot citation. So the honest grade is moderate: a documented path with no measured size.

The most cited public test is OtterlyAI's 2,000 AI blogs experiment. Otterly published about 1,000 AI-generated posts on each of two new capybara sites. Google deindexed both, but Bing never did. Otterly reports combined Bing impressions grew every month, to 119,935 in July. Of the 3,934 AI citations Otterly tracked for one site, 2,718 (69%) came from Microsoft Copilot. Microsoft's own AI Performance report credited that site with 42,817 citations, which shows how much a prompt sample can miss.

Read it carefully before acting on it:

• It is one pair of sites, so nothing separates the tested variable from domain noise.
• Capybara questions are informational and low-stakes. That says little about commercial categories with established competitors.
• There was no human-written control.
• A vendor that sells AI tracking ran it.

What it supports: Copilot can keep citing pages Bing holds, even after Google drops them. What it doesn't support: that thin volume publishing earns Copilot citations in your market. For the difference between those two readings, see correlation vs causation in AI search studies.

IndexNow lets a site notify participating search engines when a URL changes. Microsoft's February 2026 post introducing the AI Performance report says IndexNow "helps ensure that AI systems reference the most current version of a page when generating answers." That is a first-party assertion from a protocol operator. No study supports it.

The protocol itself makes no AI claim:

• The IndexNow documentation never mentions AI, LLMs or Copilot.
• An HTTP 200 response confirms only that the ping was received, not that anything was indexed.
• As of the registry fetched in August 2026, seven engines consume it: Bing, Yandex, Seznam, Naver, Yep, Internet Archive and Amazonbot. Google is not one of them.

A plausible chain exists: IndexNow speeds Bing's view of a change, and Bing backs Copilot. Each link is documented. The end-to-end effect on Copilot citations is unmeasured. Treat IndexNow as free Bing hygiene with no AI guarantee. More detail is in IndexNow and AI search.

Microsoft names freshness in its own GEO guidance. It does not say how Copilot weighs it, and no published study tests Copilot specifically.

The main supporting study, from Seer in July 2026, found 75% of cited pages had been updated within a year. It looked only at pages that were already cited. With no control group, nobody knows whether 75% of all crawlable pages carry a recent date too. If they do, the finding is no signal at all. The dates also came from publisher-declared fields, which anyone can change without changing the content.

One lab signal points the other way. The 252,000-trial factorial study found that including explicit price information and a recent timestamp helped consistently, but it ran on a synthetic corpus with pages already in context. So freshness is unresolved, not refuted.

The practical move: keep visible dates, sitemap lastmod and Last-Modified headers consistent, and update pages when facts change. Don't bump dates to chase citations. See content freshness for the definition and the evidence.

Neither has evidence as a Copilot lever.

Schema. Ahrefs tracked 1,885 pages that added schema against matched controls and found no major uplift in citations on any platform it tested. AI Overviews citations fell 4.6%. Every treated page already had heavy AI Overview citations, so whether schema helps a new page enter the answer is untested. Existing markup does no harm. It just isn't a citation lever.

E-E-A-T. This is Google's concept, and Google says it isn't a specific ranking factor. There is no score to raise, and Microsoft documents nothing equivalent for Copilot. In Otterly's test, the site with a fictional expert author box was the one Copilot cited most, which also shows author signals were not what drove those citations. Google's own position is covered in Google's helpful content guidance and E-E-A-T.

Start with access, because it is the part you control and can verify.

1. robots.txt: confirm bingbot is allowed on the pages that matter. A blanket AI block can catch search crawlers too.
2. Firewall and CDN: a WAF can return 403 to a crawler that robots.txt allows. From 15 September 2026, Cloudflare blocks Training and Agent crawlers by default on ad-displaying pages for new domains and existing free-tier customers, while Search stays allowed. Check your own settings rather than assuming. The AI Crawler Checker requests a page as each AI crawler and shows which ones get turned away.
3. Suppression tags: search templates for data-nosnippet, nosnippet, max-snippet:0 and noindex. Bing states content marked data-nosnippet is still indexed but excluded from snippets and AI summaries.
4. Bing Webmaster Tools: verify the site, confirm index status, and read the AI Performance report. Treat Citation Share as observational, as Microsoft does.
5. Facts on the page: state prices and key numbers in the HTML so a retrieved page carries a sentence Copilot can quote.

The Bing and Copilot guide walks through each step. None of this guarantees a citation. It removes the reasons you can't get one.

Does ranking well in Bing mean Copilot will cite me? Not necessarily. Bing's index backs Copilot, so being indexed is a precondition. But Microsoft publishes no data on how Bing rank translates into Copilot citations, and no outside study has measured it. Retrieval also runs on queries Copilot writes itself, not the prompt a user types. Treat Bing indexing as necessary, then check which grounding queries actually surface your pages in Bing Webmaster Tools.

Does Copilot render JavaScript? The evidence base doesn't settle it. Vercel's crawler study, which found ChatGPT and Claude crawlers fetch JavaScript but don't execute it, excluded Microsoft Copilot because it lacked a unique user agent. The safe move costs little: put prices, product names and key claims in the server-rendered HTML so any crawler reading the raw page gets them.

Is Bing Webmaster Tools' AI Performance data available by API? As of the last check in August 2026, no. The report shows total citations, cited pages, grounding queries, intents and Citation Share in the interface only, and the Bing Webmaster API documentation mentions none of them. Microsoft has indicated API access may come during 2026. Until then, export or screenshot the report to keep a history.

Will blocking GPTBot affect Copilot? GPTBot is OpenAI's training crawler and has nothing to do with Microsoft's index. Copilot depends on Bing, which crawls with bingbot. The common real-world error is a broad AI-bot block or firewall rule that also catches search crawlers. Check each user agent separately, since every vendor's controls are independent.