What AI reads before it recommends.
Ask an AI assistant which software to shortlist and the answer arrives with a list of sources. We went through more than 1.3 million of them. They come from 11 months of answers about 116 B2B technology brands, and if you're working on generative engine optimisation (GEO), or trying to get your company cited by ChatGPT, they show which pages the assistants read before they recommend anyone.
- Citations
- Over 1.3 million
- Period
- October 2025 to September 2026
- Brands
- 116 B2B technology brands
- Assistants
- Eight, mostly ChatGPT, Gemini and Perplexity
AI mostly reads what companies publish.
Company websites supply far more of what the assistants read than any other kind of source. The pages they pick follow a pattern, and a marketing team can plan around it. These six findings sum up the report; each has its own section below.
- 83%
- of citations point to company websites. Media and every other kind of source share the remaining 17%.
- 22%
- of citations go to blog posts and articles, the most-cited type of page.
- 6%
- of citations go to a brand's own site when the question doesn't mention the brand. When it does, the figure is 29%.
- 1%
- of pages earn 30% of all citations. Most pages in our data were cited once and never again.
- +36%
- more citations on average go to pages with FAQ markup. Visible dates and longer pages are linked to more citations too.
- 5 months
- is the typical age of a dated page when an assistant cites it. Four in ten were under three months old.
All figures are rounded. They come from our analysis of Agentcy data from October 2025 to September 2026.
Most citations go to company websites.
About two-thirds of all citations go to other companies' websites. In practice that means other vendors, and the service firms that write about your category. Another sixth go to the brand's own site, which puts company websites behind more than four in every five citations in our data.
Everything else comes to about 17%. That is less than many PR and SEO plans assume. Review sites and the press sit in that 17%, and because they're where the assistants go for a view from outside the vendor, we've given the media side a report of its own, Earned media in the age of AI answers.
- 66%
- of citations go to other companies' websites, including competitors and service firms in your category
- 17%
- of citations go to the brand's own website. The share grew all year.
- 2% each
- for review sites, trade media, forums, reference sites and YouTube
- 1% each
- for national press, blogging platforms, government sources and consultancies
When buyers ask about the category, someone else writes the comparison.
Blog posts are the most-cited type of page overall. Comparison pages and best-of lists tell you more. Across company websites they earn about one in ten citations, and on brands' own sites about one in thirty, so when an assistant compares you with your competitors it is most likely reading a page that another company in your category wrote.
Blog posts come first
Blog posts and articles take 22% of all citations. On brands' own sites the mix is more even, and product pages (16%) and documentation (12%) sit close behind the blog.
Documentation and PDFs get read
Help centres and developer documentation get cited a lot when buyers ask technical questions about setup, integration or security. PDFs take about 4% of citations. Most are whitepapers and reports, and they only get cited if a machine can read the text inside them.
Pricing pages and case studies are rare
Each takes about 1% of citations. We suspect many are gated, vague or missing altogether. Buyers still ask about price and proof, and the assistants look for the answers on other companies' pages instead.
- 29%
- of citations go to the brand's own site when the question names the brand
- 6%
- when it doesn't, which covers the category and problem questions where shortlists form
- 2.5%
- of comparison and best-of pages cited on those questions are the brand's own
- 31%
- of citations go to pages that three or more different assistants cite
Pages with FAQs and more depth get cited more often.
Citations bunch up. The top 1% of pages take 30% of them. We audited about 190,000 of the cited pages to see what the most-cited ones have in common, and found that pages which are easy to scan do better, while longer pages with more detail do better again.
- +36%
- more citations on average for pages with FAQ markup, 5.9 against 4.4
- +26%
- more citations for pages that show when they were last updated
- +24%
- more for pages with lists, although nine pages in ten already have them
- +16%
- more citations for pages that include a table
- 6.1 vs 3.3
- average citations for pages over 5,000 words, compared with pages under 500 words
- No change
- from article markup on its own. A named author makes only a small difference.
These are associations. They don't prove that any single feature causes more citations, because sites with the budget to add FAQs and keep their dates current usually have strong content and plenty of inbound links as well, so we'd read them as the habits of well-kept pages.
Buyers see different pages at each stage and in each assistant.
Early in a purchase, the assistants lean on best-of lists, which show a buyer who is in the market. Closer to a decision they turn to evidence such as review listings, pricing pages, case studies and PDFs. Your own site's share climbs at every step. It goes from 6% of citations at the awareness stage to 20% at the decision stage.
The assistants differ a lot in what they cite. ChatGPT draws mainly on documentation and product pages. It rarely cites best-of lists or comparisons, which make up a much larger part of what Perplexity and Gemini read, along with blog posts. Give two assistants the same question and they can build their answers from quite different pages, which changes where you should put your effort if your buyers use more than one of them.
- 7% → 4%
- is the fall in best-of lists' share of citations from awareness to decision
- 1% → 2%
- for case studies, and for pricing pages, as buyers near a choice
- 30%
- of ChatGPT's citations go to documentation and product pages, against about 10% for Perplexity
- 14%
- of Perplexity's citations go to best-of lists and comparisons, against under 2% for ChatGPT
Assistants are citing brands' own sites more often.
Over the 11 months of data, the brand's own site went from about one citation in twelve to about one in five. The rise shows up whether or not the question names the brand, which suggests it goes beyond any change in the mix of questions we asked. Comparison pages grew roughly fivefold, to about 5% of citations. PDFs went from almost nothing to about 4%.
Where we could find a publish date, most cited pages were less than a year old. The typical one was about five months old when an assistant cited it. Only 6% were more than three years old. News gets cited while the story is still fresh, whereas trade features, forum threads and company pages stay in use for many months after they go up.
- 8% → 21%
- is the own-site share, late 2025 to the third quarter of 2026
- 0.6% → 2.1%
- is YouTube's share over the same period. Video became a regular source.
- 1.0% → 2.3%
- for forums and communities, most of them threads on Reddit
- 2.7% → 1.5%
- for reference sites such as Wikipedia, which lost ground as other sources grew
How to get your pages cited by AI.
Most of these are changes to what you publish and how you lay it out, plus the habit of going back to update it. Most marketing teams can make them with the people they already have.
Write for the question that doesn't name you
Buyers build their shortlists from category questions and “best tool for” questions. Your own site gets about 6% of the citations on those. Pages that answer them directly have the most room to grow.
Publish your own comparisons
Write fair pages that compare you with named alternatives, and pages that say which customers you suit best. Otherwise the assistants build their comparisons from pages other companies wrote.
Treat documentation as marketing
Documentation earns about one in eight of the citations that go to brands' own sites. Keep it public and current. Check that search engines and AI crawlers can reach it.
Put the answer first
Open each page with a direct answer to the question it targets. Back it up with tables and clear headings, and show when the page was last updated, because an assistant finds a page laid out like that easier to quote accurately.
Go deep on fewer pages
Because 1% of pages take 30% of citations, one thorough reference page on your subject that you keep improving is likely to earn more than a series of thin posts on the same theme.
Refresh on a schedule
Four in ten cited pages were under three months old when an assistant picked them up. Review your most important pages every quarter. Show the date you last updated them, so readers and assistants can both see the page is current.
Make your evidence readable
An assistant can't read a report locked behind a form or published as images. Put your key findings on an open page as text as well.
Earn the pages you can't write
Review sites and trade media fill the gaps between vendor pages. The only way onto them is through reviews and press coverage, and in most companies that work sits with the PR team.
Read it in full.
The report runs to twelve pages. It covers where citations go, which types of page get cited at each stage and by each assistant, what the most-cited pages have in common and how the picture changed over the year. It's free, and there's no form.
Download the PDF- Pages
- 12
- Format
- PDF, 1 MB
- Data
- Oct 2025 to Sep 2026
Try it with an AI assistant. Ask what these findings mean for your own website and content.
More from the same data.
Earned media in the age of AI answers
Which trade and national media titles the assistants cite for B2B tech. Each assistant reads from a different list.
What AI gets wrong about B2B tech brands
The errors in AI answers about B2B brands and the patterns behind them. We trace whether they start on your own site or somewhere else.
The UK B2B Tech PR AI Visibility Index
Which UK B2B tech PR agencies the assistants recommend. We measure it every week against the same set of buyer questions.
AI citation statistics for 2026.
These headline numbers from the report are rounded for quoting, and every one of them comes from over 1.3 million citations in answers to buyer questions about 116 B2B technology brands between October 2025 and September 2026.
- 83%of AI citations point to company websites. Other companies' sites get 66% and the brand's own 17%.
- 22%of citations go to blog posts and articles, well ahead of product pages at 9% and documentation at 8%.
- 6%of citations go to a brand's own site when the question doesn't name the brand, and 29% when it does.
- 2.5%of the comparison and best-of pages cited on unbranded questions are on the brand's own site.
- 30%of all citations go to just 1% of cited pages, and 58% of pages are cited only once.
- +36%more citations on average go to pages with FAQ markup, and pages showing an updated date get 26% more.
- 6.1 vs 3.3average citations for pages over 5,000 words, against pages under 500 words.
- 5 monthsis the typical age of a dated page when AI cites it, and 39% are under three months old.
- 1% → 5%comparison pages' share of AI citations, from October 2025 to September 2026.
- 8% → 21%the brand's own site's share of citations, late 2025 to the third quarter of 2026.
Our blog post How to rank in ChatGPT turns these findings into a shorter, step-by-step guide for your own website.
If you're citing this research, you're welcome to use these figures with a credit and a link. Cite it as Resonance (2026), What AI reads before it recommends. https://www.resonancecrowd.com/research/what-ai-reads-before-it-recommends
Questions about AI citations.
What content do AI assistants cite most when B2B buyers ask for advice?
Company websites, by a wide margin. In Resonance's analysis of over 1.3 million citations in AI answers about 116 B2B technology brands, about 83% pointed to company websites, two-thirds of them to other companies' sites and a sixth to the brand's own. Blog posts and articles were the most-cited type of page, at 22% of citations, well ahead of product pages and documentation.
How often do AI assistants cite a company's own website?
About 17% of the citations in our data went to the brand's own site. The wording of the question changes that a lot. When the question names the brand the share is 29%, and when it doesn't it falls to 6%, which leaves your site with a small part in the category questions where buyers draw up their shortlists. The own-site share also grew over the year, from about 8% of citations in late 2025 to 21% by the third quarter of 2026.
Does FAQ schema help a page get cited by AI?
Pages with FAQ markup averaged 36% more citations than pages without it, the largest gain linked to any single feature in our audit. Pages with a visible updated date got 26% more. Lists came with a 24% gain and tables with 16%. We can't show that any of these causes the citations, because well-resourced sites tend to do all of these things and to have stronger content and more links as well.
Do longer pages get cited more often by AI?
In our data they do. Pages over 5,000 words averaged about six citations each, against about three for pages under 500 words. The most-cited 10% of pages had a median length of about 1,700 words, and they used more headings than pages that were cited only once.
How recent are the pages AI assistants cite?
Among pages with a detectable publish date, the typical page was about five months old when an assistant cited it. Four in ten were under three months old, and three in four were under a year old. News goes stale fastest. Technology news was usually cited within about 18 days of publication, while trade features and company pages stayed in use for months.
Which third-party sites do AI assistants cite for B2B technology?
Beyond company websites, the assistants cite review sites such as G2, Capterra and PeerSpot, forums such as Reddit and Hacker News, reference sources such as Wikipedia and arXiv, and YouTube. They also cite blogging platforms such as Medium and Substack, plus news and trade media. Together these make up about 17% of citations. Each assistant has its own mix. Gemini and Perplexity cite YouTube and Reddit far more than ChatGPT, which hardly cites video at all.
Do different AI assistants cite different types of content?
They do, and the gaps are big enough to change which of your pages get read. ChatGPT leans on documentation (16% of its citations) and product pages (14%), and it rarely cites best-of lists or comparison pages. Perplexity gives 29% of its citations to blog posts and about one in seven to best-of lists and comparison pages combined. Perplexity and Gemini both cite YouTube and Reddit far more than ChatGPT does.
How do I get my content cited by ChatGPT?
Publish pages that answer the questions buyers ask, especially the category questions that don't mention you by name. In our analysis of over 1.3 million AI citations, the most-cited pages put the answer first and went into real depth, and they were more likely to use FAQs and tables and to show an updated date. Publish fair comparison pages yourself, and refresh key pages every quarter. Coverage on the review sites and trade media that AI reads gets you onto the pages you can't write yourself.
What are the ranking factors for AI search?
No assistant publishes its ranking factors. The pages AI cites most do share some clear patterns, though. In our data, pages with FAQ markup averaged 36% more citations and pages with a visible updated date averaged 26% more, lists and tables were linked to smaller gains, and longer pages were cited more, while most of the pages the assistants cited were under a year old. None of this proves cause, so we'd treat these patterns as the habits of well-kept pages.
What is generative engine optimisation (GEO)?
Generative engine optimisation, sometimes called answer engine optimisation (AEO) or LLM SEO, is the work of getting your brand and content cited in answers from AI assistants such as ChatGPT and Google's AI Overviews. It overlaps with SEO, although it leans harder on third-party sources and on clear answers the assistants can quote directly. Our research shows which sources the assistants cite when buyers ask about B2B technology.
How was the research done?
Agentcy, Resonance's AI visibility platform, asked eight AI assistants, including ChatGPT and Gemini, the kinds of questions B2B buyers ask. It logged every source they cited. The data covers almost 190,000 answers to about 67,000 questions asked between October 2025 and September 2026. We sorted the cited sites and pages by type and audited about 190,000 cited pages for features such as FAQs, tables, visible dates and length.
See which pages AI reads for your category.
Tell us who you are and we'll ask ChatGPT, Gemini and Perplexity the questions your buyers ask, then send you a short snapshot of what they say and which pages they cite.
It's free, with no obligation. We run it with Agentcy, the platform behind this research.
By submitting this form, you agree to our privacy policy.