# What Copilot Cites in B2B: 4,123 AI Citations Studied

*Go to Market · Updated 2026-10-03T01:01:00+01:00 · 8 min read*

**Over 29 days, Microsoft Copilot cited 52 of Provena's 165 B2B articles 4,123 times in Bing's sampled data. Guides to a named software category took 86% of those citations and were cited at more than twice the rate of sales and go to market pages. Within software guides, regulated B2B verticals were cited and ecommerce guides were not. Page structure made no difference: cited and uncited pages had the same length, headings, tables and FAQs. Alternatives and comparison pages were never cited, and ten pages took 69% of all citations.**

Most advice about getting cited by AI search is written without data. This study uses Provena's own: Bing Webmaster Tools reports, for every page on a site, how often Microsoft Copilot and its partners cited it in an answer, and which grounding queries those answers came from. Provena publishes 165 B2B articles built on one editorial template, so the site is close to a controlled test. Length, headings, tables and FAQs barely vary from page to page, while subject and format vary a great deal. That makes it possible to see which of those Copilot rewards.

## How often did Copilot cite the site, and how concentrated was it?

Between 27 August and 24 September 2026, Bing reported 5.4K citations for the site, 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 of articles. Those citations landed on 52 of the 165 articles, 32%. The other 113 articles were not cited once in the sample.

Citation was concentrated. The single most cited article, the [insurance distribution management software guide](/blog/insurance-distribution-management-software-guide), took 11.5% of all article citations. The top five took 46.4% and the top ten took 69.3%. An AI answer engine appears to settle on a small number of pages it trusts for a topic and return to them, rather than spreading citations across every page that covers it.

## Which subjects did Copilot cite?

The clearest split is by subject. Guides to a named software category, such as eDiscovery software, MGA software, treasury management software or dealership BDC software, were cited at more than twice the rate of articles about sales and go to market, and they took 86% of the citations. This held even though sales and go to market is Provena's own field and makes up most of the site.

| Subject | Articles | Cited | Cited rate | Citations | Share of citations |
| --- | --- | --- | --- | --- | --- |
| Software category guides | 64 | 30 | 47% | 3,557 | 86.3% |
| Sales and go to market | 101 | 22 | 22% | 566 | 13.7% |

*Articles cited by Microsoft Copilot in Bing's sampled data, 27 August to 24 September 2026, by subject.*

The likeliest reason is the shape of the questions. A buyer asking an assistant how to choose a docketing system or what to look for in an underwriting workbench wants a neutral evaluation of a category, and a page that names the vendors, the selection criteria and the tradeoffs answers that directly. Questions about outbound are more often answered from large, established sales publishers, and a newer site competes with far more pages for them.

## Did the industry behind the software matter?

Yes, sharply. Within software category guides, the regulated and specialist B2B verticals were cited and the consumer facing ones were not. Every financial services guide was cited at least once. Not one of eleven ecommerce software guides was, and neither were the three field service guides.

| Software guides by industry | Articles | Cited | Citations |
| --- | --- | --- | --- |
| Insurance | 11 | 8 | 1,204 |
| Automotive dealerships | 10 | 5 | 835 |
| Legal | 9 | 4 | 714 |
| Financial services | 6 | 6 | 515 |
| Construction | 5 | 3 | 154 |
| Ecommerce | 11 | 0 | 0 |
| Field service | 3 | 0 | 0 |

*Software category guides by industry section, cited articles and citations, same window.*

This study cannot isolate why, but the pattern is consistent with competition. Ecommerce software is covered by vendors, review platforms and publishers with very large sites, so Copilot has many established sources to cite. Insurance distribution, MGA platforms, eDiscovery and dealership BDC tools have far fewer independent buyer guides, and a careful one becomes a natural citation. The practical reading is that a B2B site earns AI citations faster where independent evaluation is scarce.

## Which page formats were cited?

| Page format | Articles | Cited | Cited rate | Citations |
| --- | --- | --- | --- | --- |
| Buyer guide | 79 | 37 | 47% | 2,891 |
| Best-of list | 23 | 7 | 30% | 1,095 |
| Playbook or explainer | 39 | 7 | 18% | 136 |
| Original research | 7 | 1 | 14% | 1 |
| Alternatives page | 13 | 0 | 0% | 0 |
| Comparison page | 4 | 0 | 0% | 0 |

*Articles by format, classified from the title, same window.*

Buyer guides and best-of lists took 97% of citations between them. Thirteen alternatives pages, covering tools such as Instantly, Apollo, Gong and ZoomInfo, were never cited, and nor were four comparison pages. Those pages answer questions about named brands, where the brand's own site and large review platforms already rank, so a third party page is rarely the source an answer engine picks.

The most surprising row is original research. Provena's five dealership datasets, including the [gateway census](/blog/dealership-email-gateway-census) and the [email statistics index](/blog/car-dealership-cold-email-statistics), earned no citations in the window, and the seven research pages together earned one. First-party data can be very useful to readers who find it, but it answered few of the questions people actually put to Copilot in this period. That is a reason to publish research for links and readers, and not to expect it to drive AI citations on its own.

## Did page structure explain which pages were cited?

No. Every article on the site uses the same template: a short answer, question-led H2 sections, comparison tables, an FAQ and a source list. The medians below are near identical for cited and uncited pages, and they stay identical when the comparison is limited to software category guides alone.

| Median per article | Cited (52) | Not cited (113) |
| --- | --- | --- |
| Words | 1,572 | 1,571 |
| H2 sections | 10 | 10 |
| H2s phrased as questions | 10 | 10 |
| Tables | 2 | 2 |
| FAQ questions | 5 | 5 |
| External sources linked | 5 | 4 |
| Internal article links | 3 | 3 |

*Median structural features of cited and uncited articles.*

This is not evidence that structure is irrelevant. Every page here already has the structure answer engines are said to prefer, so the study cannot show what would happen without it. What it does show is that once a page has a clear answer, question headings, tables and an FAQ, adding more of the same does not separate it from its neighbours. Subject and competition did that.

## What did the questions look like?

Bing reports the grounding queries behind the citations. The sampled table holds 49 of the 133 distinct queries in the window, with 2,516 citations between them. The median query was seven words long, which is a question rather than a keyword. Queries phrased around evaluation, using words such as criteria, evaluation, choosing, comparing or differences, carried 41% of the sampled citations. Short category head terms of three words or fewer, such as "docketing system" or "mga software", carried 12%.

Typical examples: "choosing electronic data discovery software compliance industries", "insurance software evaluation criteria MGAs program administrators specialty products" and "auto dealership CRM real-time pipeline BDC managers". Each describes a buyer, a category and a constraint. A page that names the constraint, such as compliance, MGAs or BDC managers, in its headings and answers it in the opening lines of the section is answering the question as it was asked.

## What should a B2B marketing team do with this?

1. Write evaluation guides for the software categories your buyers research, especially where independent buyer guides are scarce. That is where the citations were.
2. Answer the constraint, not only the category. Put the buyer type and the constraint in the heading and answer it in the first two sentences of the section.
3. Do not expect alternatives or head to head comparison pages to earn AI citations against brand and review sites. Build them for readers in a buying cycle, not for answer engines.
4. Publish first-party research for links, press and sales conversations, and pair it with a buyer guide that answers the question the research informs.
5. Measure by page and by query. Bing Webmaster Tools reports AI citations per page and per grounding query, and a site wide total hides that ten pages may be doing most of the work.

Provena builds and runs this kind of research-led content alongside [outbound and outreach](/solutions/outreach) as part of its outsourced go to market team; the [case studies](/case-studies) show the outbound side in practice. More of this work sits in the [go to market category](/blog/category/go-to-market). The data behind this page is downloadable below and on the [Provena research page](/research), and the method is set out next so the numbers can be checked.

## How was this measured?

Source: Bing Webmaster Tools, AI Performance report, citation source Microsoft Copilot and partners, read for the window 27 August to 24 September 2026 (29 days) and captured on 26 September 2026. Bing states the page and query tables are a sample of overall activity that may be refined as more data is processed, so totals in the tables are lower than the 5.4K headline. Citations reported against the site's former app hostname were credited to the same article on the current hostname. Population: all 165 articles published on provena-ai.com at the time, each read from its published markdown and HTML for word count, headings, tables, FAQ questions, linked external sources and internal links. Subject and format were classified from the title and the article's section, and every article's classification is in the downloadable table. The per article, query and daily tables are also published as an open dataset on [GitHub](https://github.com/danielmcgrattan2007/b2b-ai-search-citation-study) under CC BY 4.0. Limits: one site, one answer engine, 29 days, and an observational design. Correlation here is not causation, and a site with different authority or a different template may see a different pattern.

## Which sources does this study use?

[Bing Webmaster Tools AI Performance report, provena-ai.com, 27 August to 24 September 2026](https://www.bing.com/webmasters/). Bing announces changes to its reports on the [Bing Webmaster Blog](https://blogs.bing.com/webmaster/). Article features were measured from the published pages. Aggregate and per article figures are published; no visitor or personal data is involved.

## Frequently asked questions

### What kind of B2B pages does Microsoft Copilot cite most?

In Provena's data, guides to a named software category, such as eDiscovery, MGA or treasury management software, took 86% of 4,123 sampled Copilot citations and were cited at 47% of such articles, against 22% for sales and go to market articles. Buyer guides and best-of lists together took 97% of citations.

### Does page structure affect whether AI search cites a page?

Not in this study. Cited and uncited articles had the same median length, number of H2 sections, question headings, tables and FAQ questions, because every page used one template. The result shows that extra structure does not separate pages once each has a clear answer, headings, tables and an FAQ; it does not show that structure is unnecessary.

### Do alternatives and comparison pages get cited by AI answer engines?

None of Provena's 13 alternatives pages or 4 comparison pages was cited in 29 days of Copilot data. Questions about named brands tend to be answered from the brand's own site and large review platforms, so these pages are better treated as resources for buyers already in a cycle.

### Does original research get cited by AI search?

Rarely, in this window. Provena's seven research pages, including four dealership datasets, earned one citation between them. Research earns links and press, but it answered few of the questions users put to Copilot, so it works best paired with a buyer guide that answers the related question.

### How can I see which of my pages AI search cites?

Bing Webmaster Tools has an AI Performance report that lists citations by page and by grounding query for Microsoft Copilot and its partners. Verify the site in Bing Webmaster Tools, open AI Performance, and read the Pages and Grounding Queries tables over a 30 day window.

## Sources

- [Bing Webmaster Tools AI Performance report, provena-ai.com, 27 August to 24 September 2026](https://www.bing.com/webmasters/)
- [Bing Webmaster Blog](https://blogs.bing.com/webmaster/)

---
Source: https://www.provena-ai.com/blog/ai-search-citation-study