To rank in ChatGPT, your company needs to appear on the pages it reads when buyers ask their questions, and most of those pages are company websites. We analysed over 1.3 million citations in AI answers about B2B technology. The pages cited most gave a direct answer near the top and kept their content visibly up to date. Your own site can do part of that work. The rest comes from comparison pages and trade press coverage on other people's sites.
People ask us how to get into ChatGPT's answers as if it were a new search engine with an algorithm waiting to be cracked. Before recommending anything, we wanted to see what the assistants actually read, so we measured it. Agentcy, our AI visibility platform, recorded every source cited when ChatGPT, Gemini, Perplexity and other assistants answered the questions B2B buyers ask about 116 technology brands. That came to almost 190,000 answers over 11 months. The full findings are in our free report, What AI reads before it recommends, and this post sets out what they mean for a company that wants to be recommended.
When ChatGPT searches the web for an answer, it builds its reply from a handful of pages it considers relevant. Nobody outside OpenAI can see how it ranks them. Across tens of thousands of questions, though, we can see which pages it picks again and again.
Most are company websites. About two-thirds of all citations went to other companies writing about the category, many of them competitors, and about a sixth went to the brand being asked about. Every other kind of source, review sites and the press included, made up about 17% between them.
Blog posts and articles were the most-cited type of page, with 22% of citations. ChatGPT relies more on documentation and product pages than the other assistants do, and those two page types took about 30% of its citations.
The likeliest reason is that the questions shaping your buyers' shortlists don't mention you. Ask about your company by name and the assistants cite your own site for about 29% of their sources. Ask which platform suits a particular need and that share falls to about 6%.
Shortlists form on those category questions, and the pages the assistants read to answer them mostly belong to someone else. Comparison pages and best-of lists earned about one citation in ten across all the company websites we saw, against about one in thirty on brands' own sites, and of the comparison and best-of pages cited for unbranded questions, only about 2.5% sat on the brand's own site. When ChatGPT weighs you against your competitors, it is mostly reading a page one of them wrote, or a list you've never seen.
The pages assistants keep coming back to have habits in common. We audited about 190,000 cited pages to see what separated those cited repeatedly from those cited once, and a few features stood out.
Article markup on its own made no difference. A named author made only a small one. We read these findings as the habits of well-kept pages, because a site that bothers with FAQs and dates usually has strong content and plenty of links as well. They are associations, and they don't prove cause on their own.
Citations also bunch up on a small number of pages, with the top 1% of cited pages taking 30% of all citations, which suggests that one thorough reference page on a subject will do more for you than several thin ones.
The full research, free
Twelve pages on where AI citations go, which pages win at each stage of a purchase and in each assistant, and what the most-cited pages have in common. Download the free report, with no form to fill in.
On the whole, yes. For pages with a detectable publish date, the typical page was about five months old when an assistant cited it.
How long a page stays useful depends on what kind of page it is. Tech news gets cited within a few weeks of publication and then fades, whereas trade features and company pages stay in use for many months. For most companies the useful habit is a quarterly review of the most important pages, with the date of each update shown on the page.
Much of the groundwork is shared. Well-structured pages that search engines can crawl easily help with both. The big difference is where the answer comes from.
An AI answer reads a few pages and makes the choice on the buyer's behalf, often from sources you don't control. The assistants turn to review sites and the press for a view from outside the vendor, so a good share of the work happens off your own site. Our report on earned media in the age of AI answers covers the media side. It found that trade and specialist titles supplied most of the media citations on buying questions.
The assistants also disagree about what to read. Best-of lists and comparison pages make up under 2% of ChatGPT's citations but about 14% of Perplexity's, and Gemini cites YouTube and Reddit far more than ChatGPT does. A buyer who uses more than one assistant will see different sources, and that's a good reason to build a presence on several kinds of page.
This is the list we work through with clients. None of it needs extra headcount or a specialist team.
Once the assistants do mention you, check what they say. When we checked almost 50,000 answers against verified facts, two in three got something wrong about the company. Our report on what AI gets wrong about B2B tech brands explains where those mistakes come from.
You can't pay to be one of the sources ChatGPT cites. It chooses them from what it finds on the web, so the realistic route is to publish pages worth citing and earn coverage in the publications it trusts.
There is no official list. In our data the most-cited pages answered questions directly and showed when they were last updated, and pages with FAQ markup averaged about 36% more citations than pages without it.
It varies a lot. An assistant that searches the web can find a new or improved page as soon as it has been indexed, but what the model learned in training changes more slowly, so a consistent description of your company across the web over several months helps too.
Review sites, trade media and forums are cited as sources in their own right, alongside company websites. The assistants look there for an independent view of you, so a presence on them adds to what your own site can do.
Ask it the questions your buyers ask, including the ones that don't mention you by name, and look through the sources it lists. A fuller picture across several assistants needs a monitoring tool that works at scale. We'll happily run a free AI visibility snapshot for your company.
Pick a few questions a buyer would ask before they had ever heard of you, put them to ChatGPT and Perplexity, and note which pages come back. If none of them are yours, you've found the first page to write. For the detail behind every figure in this post, download the free report, or ask us for a free snapshot of what AI reads about your category.
Figures come from Resonance analysis of Agentcy AI visibility data, which covers over 1.3 million citations in almost 190,000 AI answers about 116 B2B technology brands from October 2025 to September 2026. All figures are rounded.