Check whether Bing AI citations turned into Copilot visits

Bing Webmaster Tools now tells you which of your pages appear as sources in AI answers. It does not tell you whether anyone clicked. On Oct 13, 2026, Bing added a beta called Topics Portfolio to the AI Performance report. It groups the grounding queries where your site is cited into topics and lists queries where other sites are cited and yours is not. Search Engine Journal covered the same release and links Bing's post as its primary source.

If you connect Bing Webmaster Tools this week, you will see citation numbers with no visit numbers beside them. This post shows how to put a visit count next to them for one month, and how to read the gap without over-reading either side.

What Bing's report counts

The AI Performance report launched in public preview in February 2026. Bing says it covers Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations. Its metrics are total citations, average cited pages, grounding queries, page-level citation activity and visibility trends. None of them is a click or a visit.

Three definitions from Bing matter here:

• A citation is a source "displayed as sources in AI-generated answers during the selected time frame." Being shown is the whole event.
• Grounding queries are "the key phrases the AI used when retrieving content that was referenced in AI-generated answers." They are not what the person typed. Bing says the data shown "represents a sample of overall citation activity."
• Page-level data reflects "how often pages are cited, not page importance, ranking, or placement."

The newer Topics Portfolio adds approximate figures. Topic Size is the approximate total citations across all sites for a topic. Sites Cited and Queries are approximate counts. Opportunity volumes are shown as ranges. Bing's wording is to "treat the numbers as directional signals for planning, not as a precise ledger." It also says the topic classification is AI-generated and can be imperfect, that branded queries can show up as opportunities, and that smaller sites "may see a shorter list or may not even see this feature."

One more limit from Bing's June 2026 update: Citation Share does not "represent traffic share", and it does not expose competitor domains.

So the report answers "where am I shown?" It was not built to answer "who came?"

Source: Cloudflare, The crawl-to-click gap: Cloudflare data on AI bots, training, and referrals, 2025-08-29. Retrieved Oct 15, 2026.

What reaches your server instead

Your own traffic records give you the other half. Peeko has two separate paths for AI, and they measure different things.

Visits from Copilot. Peeko counts visits that arrive from ChatGPT, Claude, Perplexity, Copilot and Gemini from ordinary page views, by their referrer. No middleware is involved. For Copilot the referrer host is copilot.microsoft.com, as we describe in how we count visits from AI assistants by referrer. That count is a floor, not a total. Some assistants and apps send no referrer, and those visits land in Direct, where they cannot be recovered.

Crawler hits. Crawlers do not run JavaScript, so this is a separate server side path using the @usepeeko/bot-track middleware. Peeko matches the user agent against a curated list of 55 agents from 21 companies and files each recognized hit under AI answers, Search indexing, AI training or Other AI bots. Bingbot is an example of Search indexing. An AI answers hit means a page was read while an assistant was replying to somebody. It does not mean the page was cited, and it does not mean anyone clicked.

The Cloudflare chart above shows why a fetch should not be mistaken for a visit. In Cloudflare's July 2025 data, Microsoft's crawlers requested 40.7 pages for every human visit referred from Microsoft. Perplexity's ratio was 194.8 and OpenAI's was 1,091.4. Cloudflare divides crawler requests by HTML requests whose referrer names the platform, and says that because native apps can send no referrer, the ratios may overstate the true figure by an amount it does not know. The data is from 2025 and from Cloudflare's own network, so treat it as the shape of the gap, not a number for your site. Anthropic's ratio in the same table was 38,065.7.

Put the two sides next to each other for one month

Pick a full calendar month that is already over. A month gives a small site enough days to see anything, and a closed month cannot change under you. The steps below use September 2026.

1. Bing side. In Bing Webmaster Tools, open AI Performance and set the date range to Sep 1 through Sep 30, 2026, as closely as the picker allows. Write down total citations and the top cited pages. If you have Topics Portfolio, write down the topic names, with the caveat that an AI chose them. If the tab is missing, write that down too. A missing tab on a small site is a documented outcome, not a bug on your end.

2. Visit side. Read your Copilot visits for the same dates. In the Peeko dashboard, open the AI tab and the Clicks view, which counts people rather than bots. From a terminal, the sources command lists every referrer for the dates you give it:

npx @usepeeko/cli sources --site example.com --from 2026-09-01 --to 2026-09-30 --limit 100

Find the Copilot line in that list. Replace example.com with your domain. Day ranges follow the site's time zone, which is UTC until you set another one, so a hit near midnight can land on a different day than it does in Bing.

3. Crawler side. Pull the AI crawler hits for the same dates:

npx @usepeeko/cli bots --site example.com --from 2026-09-01 --to 2026-09-30

Note the counts under AI answers, Search indexing, AI training and Other AI bots. If you want the list as data, add --json. The command line and an agent that uses our MCP server return the same numbers, so you can ask your coding agent to run the same month and get the same figures.

4. Page level. Take the three pages Bing cites most. For each, write down its Copilot visits and its crawler hits. Page level is where a mismatch becomes useful, because a site total hides which pages are cited and never clicked.

You now have three columns for one month: citations from Bing, Copilot visits from your page views, and crawler hits from your server side path.

How to read the gap

The columns will disagree. These are the common patterns and what each one can and cannot tell you.

Many citations, few or no Copilot visits. This is the usual case, and it is what the definitions predict. A citation is a source shown in an answer. Many answers satisfy the person without a click. Before reading it as a failure, check two things: how many visits sit in Direct with no referrer, and whether the number is large enough to read at all. A small site with 4 Copilot visits in a month has no trend, only a count.

Citations in the report, visits arriving under a different referrer. Bing's report covers Copilot, AI summaries in Bing and partner integrations. Peeko's AI visit count recognizes five assistants by referrer. A referrer is a hostname, not a description of what the person clicked on. A visit that arrives with a plain Bing referrer is not in the AI count, and the referrer cannot say whether an AI summary was above the link. So the two lists cover different sets of surfaces, and their totals should not be forced to agree.

Copilot visits on pages Bing does not list. The report shows a sample of citation activity, so a page can be cited without appearing in your rows. A visit is a real event on your server. A missing citation row is a sampling limit on Bing's side. Trust the visit.

Crawler hits on a page, but no citation and no visit. A fetch says the page was read. Whether it was used in an answer, shown as a source or ignored is not recorded on your server. Treat the hit as evidence that the page is reachable, not that it is winning.

Citations rising across months, visits flat. This is worth watching but not acting on after one month. Bing calls the figures directional, and a rise in a sample of citations is a rise in a sample. Repeat the same month-sized check for two or three more months before you change a page.

Once a Copilot visit does arrive, you can follow it to the next step. Our post on following an AI referral to a signup without cookies covers that, and it is the part Bing's report will never show you.

What neither side can see

Some limits are worth stating plainly, because they decide what you can responsibly conclude.

The question. An answer fetch says a page was read while an assistant was replying to somebody. The question itself never reaches your server, so nobody can report it from there. Grounding queries are the closest Bing offers, and Bing describes them as phrases the AI used to retrieve content, drawn from a sample. They are not the person's words, and they cannot be joined to a specific visit.

Why a rival was cited. Opportunity queries list where other sites are cited and yours is not. They show that the gap exists. They do not say why. Peeko cannot see inside Bing's report, and nothing in your own traffic records explains another site's citation. Bing's own caveat applies too: branded queries can show up as opportunities, so a query that names a competitor is not a content gap you can close.

Whose crawler it was. A crawler hit carries a user agent, and the user agent is a claim, not proof. For how far you can go in naming an operator without storing addresses, see naming the operator of an AI agent without IP addresses. On the crawler path Peeko stores no address and no visitor hash at all, and keeps the raw user agent up to 512 characters so old hits can be reclassified as the crawler list grows.

Anything outside Bing. Bing's report covers Bing's own AI surfaces. Your ChatGPT, Claude, Perplexity and Gemini visits will never appear in it, and your Copilot visits are only a floor. If your site is new, what AI crawlers do in the first 30 days is a useful baseline for what to expect before any citation exists.

What to do with the result

Write down the month in one line per page: citations, Copilot visits, crawler hits. Three outcomes are possible, and each has a plain next step.

• Cited and visited. The page works as an answer source and as a destination. Leave it alone and make sure the visit has somewhere to go next.
• Cited, never visited. The answer may be complete without a click. Look at whether the page offers something an answer cannot, such as a tool, a file or a number the reader needs in full. Do not rewrite it based on a directional estimate alone.
• Visited, never cited in your rows. Bing shows a sample, so absence is weak evidence. The visit is the stronger signal.

Repeat the same check next month with the same steps, so you compare like with like. If Topics Portfolio never appears for your site, you lose the topic view but not the comparison: the citations and pages view, the referrer count and the crawler hits are still enough to tell a citation from a visitor.