Copilot Cites Category Guides, Not Comparison Pages: A B2B Citation Study

Most B2B GEO playbooks assume comparison pages and "alternatives to X" content are the fastest path to AI citations. A new first-party study from Provena, published on 3 October 2026, suggests the opposite for Microsoft Copilot: alternatives and comparison pages were never cited once, while category guides to named software products took 86% of all citations.

The finding comes from Bing Webmaster Tools' AI Performance report — the first dataset that lets a publisher see exactly which pages Copilot and partner AI experiences cite, and which grounding queries triggered retrieval.

What Provena measured

Between 27 August and 24 September 2026, Provena founder Daniel McGrattan tracked citation activity for all 165 articles published on provena-ai.com. Bing reported 5.4K citations over the window, rising from 54 on the first day to a peak of 367 on 23 September.

The per-page table Bing exposes is a sample of that activity and accounts for 4,123 citations across 52 of 165 articles — 32% of the library. The other 113 articles were not cited once in the sample.

Every article was classified by subject and format from its title and section. McGrattan published the per-article, per-query, and daily tables as an open dataset on GitHub under CC BY 4.0.

Limits are explicit: one site, one answer engine (Microsoft Copilot and partners), 29 days, observational design. Correlation is not causation. A site with different authority or a different template may see a different pattern.

Finding 1: Category guides dominate; comparison pages are invisible

Of the 52 cited articles, guides to a named software category — "best CRM for small business," "top project management tools for agencies" — accounted for 86% of citations.

Alternatives pages ("X vs Y," "alternatives to Z") and head-to-head comparison content were never cited in the 29-day window — despite being part of the same site, the same template, and the same topical authority.

This contradicts the prevailing B2B GEO assumption that comparison content is citation bait. For Copilot, the pattern looks more like category-level evidence than vendor-level debate.

Finding 2: Citations concentrate on a tiny fraction of pages

Citation activity was heavily skewed. Ten pages took 64% of all citations in the sample. The remaining 42 cited pages shared the other 36%.

This matches the broader pattern BrightEdge documented in February 2026: when AI citation sources change, they change binary — domains drop out entirely rather than fading gradually. See AI Citation Volatility: Why Rank-Style Tracking Fails.

Finding 3: On-page structure did not differentiate cited from uncited

Provena compared cited and uncited articles on measurable on-page features:

Within this single-site dataset, format and topic mattered more than structural GEO tactics. A page about a named software category got cited; a comparison page on the same site did not — regardless of headings, tables, or FAQ markup.

That does not mean on-page structure is irrelevant everywhere. Indexably's cross-site study of 18,129 pages found page signals lift citation odds in the top authority quartile — see Page Optimization Only Helps High-Authority Domains. Provena's result says that within one B2B site at one authority level, content type was the dividing line.

Finding 4: Regulated verticals follow the same pattern

Provena operates in regulated B2B verticals (financial services compliance tooling). The category-guide-over-comparison pattern held across subjects — cited and uncited articles in regulated topics showed the same structural profiles as uncited articles in less regulated ones.

Regulation did not force Copilot toward different source types. Topic framing did.

What this means for B2B marketers

Rethink the comparison-page sprint

If your Copilot strategy is "publish 20 alternatives pages this quarter," Provena's data offers no evidence that Copilot will cite them. Category guides that name the software landscape — not head-to-head vendor battles — earned nearly all citations.

That does not mean comparison pages are worthless for SEO or conversion. It means they may not be the Copilot citation lever many teams assume.

Index in Bing and monitor AI Performance

Copilot retrieves through Bing. If your site is not indexed in Bing, or blocks Bing's crawlers, Copilot cannot cite you. Provena's entire dataset came from Bing Webmaster Tools' AI Performance report — the only publisher-facing tool that shows grounding queries and per-page citation counts for Copilot.

ChatGPT also retrieves through Bing for web-grounded answers, so Bing indexing matters for both surfaces — see Why Each AI Engine Reads a Different Search Index.

Expect extreme concentration

If you get cited, you will likely get cited a lot on a handful of pages — or not at all. Reporting "we published 40 new pages" without checking which pages Copilot actually retrieves is measuring output, not visibility.

Do not assume Copilot equals ChatGPT

Kern Media's parallel study of 1,044 AI answers found ChatGPT cited company websites in 82–97% of cases, while Perplexity cited company sites only 50% of the time. Copilot's B2B pattern — category guides, zero comparison citations — may not transfer to other engines. Engine-specific measurement is not optional.

How Obsurfable fits

Obsurfable records how AI systems answer buyer prompts — which brands they name, which sources they cite, and how those answers change over time. Provena's study shows what one publisher can learn from Bing's AI Performance report on their own site. Obsurfable extends that to competitive prompts: whether your category guides get cited when a buyer asks Copilot or ChatGPT for recommendations — and which competitors' pages take the slots yours do not.

The Visibility Director flags prompt-level gaps and drafts content aimed at the retrieval layer that is actually failing.

Should we stop publishing comparison pages entirely?

No. Comparison pages may still rank in traditional search, convert visitors, and support sales. Provena's finding is specific to Copilot citations over 29 days on one B2B site. It says comparison pages were not the citation path there — not that they have no value anywhere.

Does this apply to ChatGPT?

Partially. ChatGPT retrieves through Bing for web-grounded answers, so Bing-indexed content is eligible. But ChatGPT's citation behavior differs — it averages ~5 sources per answer versus Copilot's patterns in this dataset, and it may weight sources differently. Measure both.

How do I check if Copilot cites my site?

Verify your site in Bing Webmaster Tools and open the AI Performance report. You will see citation counts, grounding queries, and per-page breakdowns. If you see zero activity, check Bing indexing status and whether your content blocks AI crawlers.

Why would Copilot prefer category guides over comparisons?

Speculative, but consistent with how retrieval systems work: category guides aggregate evidence about a software class, while comparison pages argue between two vendors. A system building a neutral recommendation may prefer the broader framing. Provena does not test the mechanism — only the outcome.

Bottom line

On one B2B site over 29 days, Microsoft Copilot cited 32% of articles and ignored 68%. Category guides to named software took 86% of citations. Alternatives and comparison pages took zero. On-page structure did not separate cited from uncited pages. Ten pages captured 64% of all citations.

For B2B teams optimizing for Copilot: index in Bing, monitor AI Performance, invest in category-level guides, and measure whether your comparison-page sprint is producing citations — because on this evidence, it is not.