How ChatGPT Searches Your Website and What It Looks For
When ChatGPT goes looking for a specific fact about a brand, it goes to the brand's own website more than anywhere else. In Ahrefs' study of 15 million ChatGPT fan-out queries, about three out of four site: searches aimed at the 160 or so domains it looked up most often pointed at a company's own site, and the words it leaned on more and more in September were "official" and "price" (Ahrefs).
That's a different picture than the one a lot of us have been working from for the past year, where most of the advice was about getting into Reddit threads, Wikipedia, listicles and review sites. Those still matter, and I'll get to them. Still, if you want to know how ChatGPT searches your website, the short answer is that it does it directly, with the site: operator, and the words it adds to those searches point at prices, specs and official details. Whether those facts are sitting on your site in plain text decides a lot of what happens next.
The short version
Among the sites ChatGPT searches most with the site: operator, 75.9% of those searches went to a brand's own website, per Ahrefs.
"Official" showed up in 13.1% of English fan-out queries in September, and "price" rose by about 47% in a single month.
On factual questions about pricing, features or company details, 63.4% of ChatGPT's citations went to official websites. On comparison questions that fell to 46.4%.
Your own pages carry the factual questions, and other people's pages still carry most comparisons, so you need both.
What Ahrefs looked at
A little background on the mechanics first, since the term gets thrown around a lot. When you ask ChatGPT something and it decides to search, it doesn't always run your words as typed. It breaks the question into smaller searches, which people call fan-out queries, and runs those against the web. Ahrefs pulled 66.6 million free, logged-out ChatGPT responses from August and September 2026 and studied the 15 million fan-out queries inside them (Ahrefs).
A couple of numbers are useful to have in your head before the brand stuff. Only 14.4% of those responses contained fan-out queries at all, and the ones that did averaged about 1.6 each. (Ahrefs first described the 14.4% as the share of responses that searched the web and later corrected the wording, which I appreciate a research team owning in public.) So most answers in this sample came without a visible search, and when ChatGPT did search, it ran a small handful of queries rather than dozens. Free, logged-out sessions search less than paid and reasoning models do, so treat these as the floor.
Of all those fan-out queries, 6.8% used the site: operator to search a single website, spread across roughly 116,000 domains. That's the part brand people should care about.
How ChatGPT searches your website
Among the roughly 160 sites ChatGPT searched most this way, 75.9% of the site: searches went to a brand site, meaning a company's own website like apple.com or ikea.com. Retailers like Walmart, Costco and Amazon made up 11.9%, and Reddit and Facebook together were 1.9% (Ahrefs). Apple, IKEA and Nike sat near the top of the list, along with government sites like gov.uk and the FDA.
Ahrefs, 15 million ChatGPT fan-out queries
When ChatGPT searches one site, it's usually the brand's own
Share of site: searches among the ~160 domains ChatGPT searched most often, Aug to Sept 2026.
Source: Ahrefs, How ChatGPT Does Query Fan-Out (Oct 2026), free logged-out ChatGPT responses, Aug to Sept 2026. "Everything else" is the remainder after the three named groups.
So when ChatGPT decides it needs something specific about you, it goes to your site and looks around. Ahrefs' own advice off the back of this is that those facts have to live on your site, in plain text, on a page the engine can find.
The words it searches with tell you what it wants once it gets there. "Official" appeared in 13.1% of English fan-out queries in September, up from 10.1% in August. "Price" went from about 6% to about 8.9% of fan-outs over the same month, a relative jump of about 47%. "Reviews" grew by about a third, and "specifications" made the top 50 terms for the first time, while "meaning" and "definition" each fell by about a fifth (Ahrefs). That's one model over two months, so I wouldn't build a forecast on the trend line. The direction does line up with the questions buyers ask right before they decide, which is why I'm paying attention to it.
Your site answers the facts, and other sites answer the comparisons
Ahrefs also ran a separate test with 20,000 US prompts. On factual questions about pricing, product features or company details, 84.4% of answers cited at least one official source, and 63.4% of all the citations went to official websites. On comparison questions (think "A vs B" or "which is cheaper"), 68.3% cited at least one official source, and official sites got 46.4% of citations, so most of the sourcing came from somewhere else (Ahrefs).
Who gets cited depends on the question
On factual questions, official sites get most citations
Pick a question type to see where the citations went.
The buyer asks
Searches it might run (illustration)
Citations to official websites
That lines up pretty closely with what I saw in my own AI Verdict Study, which used a completely different setup (API calls to four engines, a seven-question buyer script, 32,089 citations). On the first, unbranded question, the brand's own site supplied 4% of the classified citations. Once the buyer got to tradeoffs and fit, the brand's own site jumped to 35% and 37%, and help center and support docs made up 46% of the brand-site citations on the tradeoffs question (Where AI Engines Go to Decide About Your Brand). Two very different studies came out with the same basic shape. Broad, early questions go out to the rest of the web, and specific "does it do this, what does it cost" questions come back to you.
My guess is a lot of brands spent the last year working on the first half of that and neglecting the second. If the engine comes to your site looking for a price and leaves without one, it's going to get a price from somewhere, whether that's a reseller, a comparison site or an old forum thread. I can't prove that from the Ahrefs data (they didn't track what happened after a site: search came up empty), so call it a hunch, though it's a hunch based on how these systems behave whenever they can't find what they're after.
Google is heading the same way
A few days after the Ahrefs study went up, Google announced five changes to AI Overviews and AI Mode meant to send more people to the web. One of them is a test that puts official sources near the top of some AI responses for task-based searches, because, as Google put it, for some tasks there are "only one or two official places on the web" (Google). I wrote about the inline-links part of that announcement in How to Get Clicks From AI Overviews and AI Mode. The official-sources test is the part that connects here. It's only a test so far, though both companies seem to be leaning toward sending factual, task-type questions to whichever site owns the answer.
A note on Reddit and Wikipedia
The same Ahrefs piece shows ChatGPT's Reddit citations fell 88% over six months, from about 1 in 5 answers in March to about 1 in 43 by late September, and Wikipedia went from 1 in 7 to 1 in 35 (Ahrefs). I covered the Reddit side in Is Reddit Worth It for AI Visibility? and the churn side in How Long Do ChatGPT Citations Last?. I wouldn't walk away from either one. The comparison numbers above say other people's pages still carry most of the weight once a buyer asks "A or B," and ChatGPT's sourcing has swung hard enough this year that I'd expect it to swing again.
What I'd do with this
Number one, run the searches yourself. Type site:yourbrand.com price, site:yourbrand.com specifications, and site:yourbrand.com plus the name of your top product into Google or Bing and look at what comes back. It's a rough proxy, since ChatGPT doesn't necessarily see the same results you do, but if you can't find your own pricing that way, I wouldn't bet on a fan-out query finding it either.
Two, put the facts in plain text. I'd push hard on this one for B2B brands that hide pricing behind "contact us," and for anybody whose specs live inside a PDF or an image. If you can't publish a price, publish a range, a starting point, or what drives the price. Give the engine something it can quote.
Three, treat your help center and support docs as marketing pages, because the engines already do. Make sure they aren't behind a login, make sure crawlers can reach them, and make sure they say plainly what the product does and doesn't do. If you've blocked AI crawlers from your docs at some point, it's a good time to revisit that call (I went through the tradeoffs in Should You Let AI Read Your Site for Free?).
Four, label the official stuff as official. If you have an official pricing page, spec sheet or returns policy, say so in the page title and the copy. I don't know how much weight "official" carries as a matching term (nobody outside OpenAI does), but it costs nothing, and it's how ChatGPT is wording a growing share of its searches.
Then last, keep working the comparison side. Your own site won't win the "A vs B" question by itself. Review coverage, comparison pages, analyst write-ups and customers talking about you in public are still most of what the engine reads for that kind of question.
Self-check, 6 questions
Can ChatGPT find your facts on your site?
Answer for your own brand. Your score and a short fix list show up when you finish.
0 of 6 answered
If you want a quick read on whether the engines can find your facts right now, the free Quick Check is a good place to start, and the $199 AI Visibility Audit Kit walks you through checking all of this yourself, page by page. For the bigger picture on why showing up inside the answer is the number to watch, read The Click That Didn't Happen. I go into why engines trust some sources over others in my book, Explainable.
I'd love to hear what turns up when you run those site: searches on your own brand, especially if the engines are pulling your prices from somewhere you didn't expect. Send me a note through the site and tell me what you found.
Jarred Smith is the author of Explainable: Why AI Recommends Some Brands & Ignores Others, an Amazon bestseller on AEO, GEO, and SEO. He's a marketing leader with nearly 20 years of experience across healthcare, public media, retail, and environmental services. Find him at jarredsmith.com.