88% of 27.3K Copilot Citations Landed on One URL

Intro

Citability.dev is an AI-citation measurement product I build and operate. In August 2026 its own Bing Webmaster Tools AI Performance dashboard recorded a step change I had not engineered for, which made it the best available case study in a field where almost nobody has first-party data.

Goal

Understand which URLs Microsoft's Copilot surfaces actually select as sources, and whether citation behaviour can be measured precisely enough to sell an intervention against.

Process

Instrumented the site against Bing Webmaster Tools AI Performance, cross-referenced against Google Search Console, and built citation tracking into the product itself so the measurement is repeatable rather than a one-off screenshot. Treated the event as a longitudinal study: citation volume, cited-page count, and query intent tracked over time rather than reported once.

Result

27.3K total citations between 19 May and 17 August 2026, sourced to Microsoft Copilots and Partners. The distribution is the result, not the total. Eleven URLs were cited at all. One of them, citability.dev/blog/tools-that-measure-domain-citation-rate, took 24.1K of the 27.3K. The second took 2.9K. The remaining nine share 232 between them. The daily series is flat through 6 August, 267 citations across the preceding 80 days, then steps to 2.3K on 7 August and has held between 2.2K and 2.7K every day since, peaking at 2.7K across 8 URLs on 17 August. Across 26 grounding queries the top four carry 75% of the volume, led by "platforms comprehensive citation rate analytics" at 8.4K citations and 32.04% citation share.

What I take from it

I have not established causation and I will not claim it. The commercially useful finding is structural: Copilot selects individual URLs, not domains, and the concentration is extreme enough that a single page can account for nearly nine citations in ten. This is first-party evidence from one engine. I have not measured whether ChatGPT, Perplexity or Gemini concentrate the same way, and I will not imply that I have. It is still enough to show that an audit reporting one site-level AI visibility score is measuring the wrong unit. Microsoft labels this dashboard's data a sample of overall activity, which is worth stating plainly to anyone reading the numbers.