Bing AI citation reports: separate newly cited pages from changed results

Separate newly cited pages from results on existing pages

Bing’s AI Performance report can help you see where your pages are cited in AI answers. A rising total needs a second question: did the same pages receive more citations, or did more pages enter the report?

Bing is Microsoft’s search engine, and Bing Webmaster Tools is its reporting service for site owners. Microsoft’s February 10, 2026 public-preview announcement describes citations across Copilot, Bing AI summaries and selected partners. Copilot is Microsoft’s AI assistant. The report’s citation measures should not be read as a universal ranking or a count of readers who visited your site. (Bing AI Performance public preview announcement ↗)

Citation totals can rise while existing pages lose ground

Illustration: ten existing guides receive 100 citations in period A. In period B those same guides receive 80, while ten newly added guides receive 60. The total rises to 140, up 40%, even though the original guides fell 20%.

Keep disappearing pages in the comparison

Freeze the list of pages in period A, then look for every one of them in period B. Do not restrict the comparison to URLs appearing in both exports: that would discard original pages that disappeared. Add newly published guides as a separate group. Check redirects or renamed URLs before treating a page as missing.

For an absent row, determine whether the report establishes zero citations or whether reporting coverage is unknown. Keep unknowns visible; do not silently replace them with zeros or drop them. The example’s 20% decline assumes complete, comparable counts for all ten original guides. If coverage is incomplete, report the observed subtotal and missing pages rather than claiming an exact group-wide change.

Use equally long periods and consistent report definitions. Inspect a declining page’s actual answer, evidence and date before choosing a repair; the count alone does not explain the cause. Compare visits and paid starts separately because citation totals are not a matching conversion denominator.

Compare the pages that disappeared as well as those that gained

Export the earlier set of pages before choosing the comparison period. Label each as still visible, absent from the later report, or changed to a new address. Investigate the changed addresses before counting them as losses. Map redirected URLs to their original pages.

In the editor’s note, give both the total change and the change among the original comparable pages. Then select one gaining and one declining page for inspection. Record whether their subject or content changed during the interval. That does not isolate a cause, but it gives you a better next investigation than treating the headline total as a reason to rewrite every cited article.

Show the effect of an incomplete comparison

Suppose ten pages had 100 citations last month. This month eight of those pages still appear and total 90 citations. The remaining two are absent from the report. You cannot conclude that the original group gained citations by comparing the 90 with just the eight pages’ previous totals and silently dropping the other two. Report the comparable subset and label what you know about the missing pages.

If the two URLs were replaced during a migration, reconcile them before calling the disappearance a loss. If you have no evidence that the report treats an absent row as zero, show an unknown separately. This makes the analysis useful even when the dashboard cannot answer every part of the question.

How to reconcile citations when a story moves

Suppose four pages received 60, 20, 15 and 5 citations in the first period. The second export lists 48 for the first page, 30 for a replacement address that you confirm serves the second article, and 12 for the third. The fourth is absent. Compare the three reconciled articles first. They went from 95 to 90, a decline of about 5.3%. The absent article makes the full-cohort result unknown unless you can verify what that absence means.

Now suppose a new article earned 25 citations. The second export shows 115 citations across the rows in this example, up from 100 in the first period. That headline gain and the comparable group’s decline can both be accurate. Show both figures and name the coverage difference. If the fourth article really had zero, the original group would have fallen from 100 to 90. If the row is missing for another reason, treating it as zero overstates what you know.

Use a small reconciliation sheet with an article ID, both URLs, both counts, and a status such as matched, moved, new, or unknown. Verify moved pages by checking that they still answer the same question. A redirect to a general category is a different case from a replacement URL for the same reporting. Keep that difference visible in the comparison.

The editorial decision changes with the result. If the increase comes from a new article, examine why that article answers a question the older set misses. If matched articles lose visibility, inspect specific query and page pairs. Neither investigation starts by rewriting the entire site. Record reader visits separately if the publisher wants to know whether the change mattered to its audience.