What 5,600 Copilot citations to an 86-post blog look like

A guide we published on 2 August had no recorded citation from Microsoft Copilot in the 42 days that followed (Bing reported on 34 of them). Then it was cited 254 times in 33 days, 216 of them in 17. It is one of 26 posts, out of 86 on the blog, that Bing Webmaster Tools recorded citations for between 17 July and 15 October 2026, in a report Bing calls a sample. Those 5,608 citations went 79% to two pages, and the weekly total barely moved while the pages under it started, peaked and stopped.

One small site, one report, one family of surfaces: a case to learn from, not a benchmark.

Where the data comes from

Bing Webmaster Tools' AI Performance report counts citations from what Bing groups as Microsoft Copilot and partner surfaces; we write Copilot for short. We read its rolling three-month view once, on 17 October 2026, covering 17 July to 15 October (91 days, 13 weeks), after a totals-only read on 3 October. It gives a daily series, a table of citations by page, and the grounding queries (Bing's term for the searches an answer was built from); clicking a page row filters the chart to that page, which is where the per-page series come from. We captured all 27 cited URLs. One is an old slug for a post already in the list, so 26 distinct posts were cited and 60 were not. Download the data: daily series and per-post table.

Bing labels the report "a sample of overall activity", so every count below is a count of the sample. What it cannot show:

• Readers. We treat a citation as one appearance of a page as a source. The report does not say who saw the answer or clicked.
• Other assistants. Nothing here is about how ChatGPT, Gemini or Perplexity behave.
• Clean totals. The 27 per-page series sum to 5,228, 380 fewer than the daily series, and the per-page query tables cover only part of each page's total. Bing's own note: "Chart totals and table results may differ when filters are applied." Where two figures disagree we quote the lower share.
• Gaps. Twenty of the 91 days read zero. We treat 27 July to 6 August and 21 to 23 August as reporting gaps rather than drops, because the pages were live throughout and some were cited on both sides (the top page went 76, gap, 50). A date next to a gap is a bound: "first cited 7 August" means on or before. The first six zeros precede the first citation.

Want the same read for your own site, across every engine we run rather than Copilot alone? Get a free visibility audit.

Finding 1: a page can sit at zero for six weeks and then surge

The agency-selection guide, published 2 August, had no recorded citation in the 42 days after publication, 34 of which Bing reported on. From 13 September it picked up 38 in 16 days, then 216 in the 17 days from 29 September, cited on 16 of them. By 17 October it was the site's third most-cited page, with 254.

It was not alone in starting late. Twenty of the 26 cited posts were published inside the window, so the lag to first recorded citation is observable: 1, 2, 5, 9, 9, 14, 19, 19, 21, 21, 42, 44, 45, 46, 46, 59, 60, 64, 66 and 81 days, median 31.5. Ten were first cited within three weeks and ten waited six weeks or more, with nothing between. Two cautions: these are the survivors (60 posts had no citation at all in the sample, 14 of them under three weeks old on 15 October, and a post published after early September cannot yet show a long lag), and most late starters have one to five citations in total, so their first citation is a single appearance. The six posts published before the window were first cited between 23 July and 1 October.

The half-life studies discussed below start their clocks at a page's first citation or its peak, so a six-week wait before the first citation is outside what they measure.

Finding 2: two pages took 79% of the citations

Of 5,608 citations, 4,449 went to two pages: a "best tools" comparison (3,493) and an alternatives guide (956). That is 79% of the daily total and 85% of the by-page sum. Thirteen of the 27 URLs got five or fewer citations in 13 weeks and 14 were cited on a single day.

Finding 3: the weekly total was steady while the pages under it moved

Eight gap-free full weeks between 10 August and 11 October ran between 501 and 656 citations, mean 554, coefficient of variation 8.9% (the week of 17 August, 567 with three gap days, is excluded, and the final four-day week, 453, is not a full week). The flatness is partly a matter of which days are in: 13 to 20 August averaged 128 citations a day, September 75. Daily counts from 24 August ranged from 22 to 175 (median 70, coefficient of variation 43%).

The per-page series show what the flat line was made of:

• The best-tools page was first recorded on 7 August, the first day after the long gap, then held 331 to 512 a week from late August, 63% to 78% of the site total in full weeks. Its biggest day was the last in the window, 158 on 15 October.
• The alternatives guide got 589 of its 956 citations, 62%, in the nine days to 20 August, when it peaked at 119, the last day before the 21 to 23 August gap. In the 56 days after (53 of them reported), it was cited on 27 days and never more than 20 times.
• The Norway market guide, published 24 July, was cited the next day, took 140 of its 207 citations between 8 and 29 September, and had none in the 16 days after. We do not know why; we will report what the next reading shows.

A steady total in a sample is not proof of a steady Copilot; Bing's sampling could be doing some of the smoothing, and that ceiling applies to every count here.

What we think it means

This section is interpretation, not measurement.

• A page with nothing in its first month has not necessarily failed. The agency guide waited 42 days; nine other posts waited six weeks or more, though most then got only a handful.
• The site line hides the churn. A weekly site-level count, which is what our earlier case study reported, can stay flat while pages swap in and out under it.
• Refreshes do not explain the agency or Norway timing. Neither page was edited after publication (its dateModified equals its datePublished). For the two older pages we cannot tell, because dateModified keeps only the latest edit.

How this relates to the half-life studies

Scrunch, with Stacker, reports that cohorts of sources that appeared in a given week lost half their citations in about 4.5 weeks (3.5 million citation events, September 2025 to March 2026, five engines not including Copilot). Profound reports that the median page falls to half its peak citation share within 11 days of peaking (883,000 pages, seven engines including Copilot, September 2025 to September 2026). Both are summarized in our citation decay guide.

We did not compute a half-life, and our numbers are not comparable: Profound measures share and Scrunch a cohort's citation activity, both across many sites; we have raw counts for one site in a sample. What we can say is that our pages do not share one shape: one burst and faded, one has held a weekly band for seven weeks, one stopped after nine weeks, one started after six. That is an illustration, not a rate, and consistent with both studies' point: a per-page citation number is a snapshot.

How to measure your own

1. Verify your site in Bing Webmaster Tools, open AI Performance, and copy the daily series and page table somewhere you own; the rolling three-month window drops older days.
2. Click each page row that matters. The filtered chart is that page's daily series; it lists only days with citations, so write in the zeros and mark known gaps.
3. Compute days from publish to first citation, the peak day, and the share of the total that came in the best week, and record the date you read it. Our 3 October read showed the Norway page at 207; on 17 October it had stopped.

Our earlier Copilot case study covers what the cited pages have in common. These numbers are from a read on 17 October 2026 and we will refresh them.