Written by Jeremy Souffir Founder, JTS Tech Services
The direct answer first. AI answer engines are leaning harder on official sources, and for a store or a B2B brand the official source is your own website. When ChatGPT decides it needs to check something about a brand, it increasingly searches that brand's site directly, and the things it is checking more often than a month ago are prices, specifications and anything it can call official. Google is moving the same way and says so. That is good news if your site states its facts plainly, and a quiet problem if your price lives in a script, your specs live in a PDF and your returns policy lives in an image. The AI arrives, finds nothing it can read, and cites someone else.
What did the study actually find?
Ahrefs, the SEO software company, analysed 66.6 million ChatGPT responses from August and September 2026 and the 15 million follow-up searches ChatGPT ran behind them, the "fan-out" queries it fires off when it decides to look something up. It also tracked which sites US answers cited from March to September. The findings that matter for a business:
- Most answers never search at all. Only 14.4% of responses triggered a web search, averaging about 1.6 searches each. Everything else came from what the model already knew. We wrote in August about what that means for the pages it does read.
- When it searches one website, it is usually the brand. 6.8% of fan-out searches were pointed at a single site using the site: operator. Among the roughly 160 domains searched that way most often, 75.9% of those searches went to brand sites, 11.9% to retailers such as Walmart, Costco and Amazon, and just 1.9% to Reddit and Facebook combined.
- It is checking prices and facts more. Between August and September the word "price" went from 6.0% to 8.9% of English fan-out searches, and "official" from 10.1% to 13.1%. "Specifications" entered the top 50 search terms for the first time.
- Official sources dominate fact questions. In a test of 20,000 US prompts, 84.4% of answers to factual questions cited at least one official source. For comparison questions (this versus that) it was 68.3%.
- The forums lost ground fast. Reddit went from appearing in about one US answer in five in March to about one in 43 by late September. Wikipedia's citations fell by around 80%.
Whose numbers these are, and where they stop
This is Ahrefs' study, and Ahrefs sells the citation-tracking tool that supplied part of the data, so read the recommendations with that in mind. It covers free, logged-out ChatGPT sessions on one model over two months; Ahrefs itself notes that paid and reasoning models search more often, so your best customers may see different behaviour. Citation data is US-only and measures citations, not clicks or sales. "Brand site" and "official" are defined by example rather than a published rule. And a shift in what gets cited is evidence of a preference, not proof of why. Treat it as a strong signal of direction, not a law.
What is Google doing?
On 15 October Google published a set of changes to AI Overviews and AI Mode. Most are about showing more links: inline links appearing higher in answers and more often, larger link previews, more news carousels. The one worth reading twice is a test. When there are "only one or two official places on the web to complete a specific task or get the exact info you need", Google says it will test a dedicated section at the top of the answer that surfaces those links, with context on why each one is useful.
Google has not said what triggers it, which queries qualify or where the test is running, so nobody can tell you yet whether a question about your store's warranty or your product's dimensions will get that box. But the intent is plain, and it matches what Ahrefs measured at ChatGPT: when the answer is a fact that one organisation owns, both systems want to point at that organisation.
Why does this matter more for a store than for a publisher?
Because a store owns a lot of facts nobody else can state authoritatively. Your current price. Whether the medium is in stock. What the jacket is made of. How long delivery takes to Alberta. Whether you take returns on sale items. A blogger can review your product, but only you can say what it costs today and what your policy is. Those are exactly the questions an AI hands off to an official source, and they are the kind of fan-out searches that are growing.
The catch is that an AI checking your site does not see what your customers see. As we covered in September, agents read the structure of a page, much as a screen reader does. A price drawn in by a script after the page loads, a size chart that is an image, a returns policy that is a PDF, a spec table that only appears after a click: a person sees all of these, and a quick automated check may see none of them. When the official source turns up empty, the AI falls back to whoever did state the fact, which may be a reseller with last year's price or a review site with the wrong dimensions.
What should you check this week?
- Run the AI's search yourself. Type site:yourdomain.com followed by a product name and the word price, then the same with returns, shipping and the main spec your buyers ask about. If your own pages do not come back with the answer in the snippet, a machine doing the same search will struggle too. A site: search can only return pages that search engines have already indexed, so check that your product and policy pages are indexed and not blocked.
- Put the facts in the page, in text. Price, availability, materials, dimensions and compatibility belong in the HTML of the product page, per variant where they differ, not only in images, tabs that load later or downloadable PDFs.
- Give every policy one plain page. Shipping times by region, returns windows, warranty terms and exclusions, written as text on a dedicated URL, linked from the footer and from product pages. These are the "exact info" questions Google's test is aimed at.
- Make the facts agree everywhere. If your product feed, your marketplace listings and your site disagree about price or stock, an AI comparing sources will notice, and you will not get to choose which version it trusts.
- Use the word official where it is true. If you are the manufacturer, say so on the page. If you are an authorised dealer, say that plainly too. "Official" is a term the AI is searching for more, and you cannot be found for a claim you never make.
The two conclusions that both get this wrong
Reddit is out, so reviews no longer matter
Not for comparison questions. The same study found official sources in 68.3% of comparison answers, which leaves plenty of room for third parties, and Ahrefs' own advice for review-heavy categories such as running shoes is to keep earning independent reviews and comparisons. The shift is that the AI goes to you for facts and to others for opinions. You need both, and only one of them is fully in your control.
It only site-searches Apple and Nike, so a smaller brand can ignore this
The biggest brands top the list because they are asked about most, not because the behaviour is reserved for them. The site: searches in the study spread across roughly 116,000 domains. And a smaller brand has fewer other places stating its facts, which makes its own site matter more, not less. If the AI cannot read your price on your site, there may be nowhere else reliable for it to look.
The genuinely encouraging part
For a long time AI visibility advice has meant chasing mentions on sites you don't control. This is the opposite. The behaviour both companies are describing rewards the one website you fully own, and the fixes are ordinary: text instead of images, one clear page per policy, consistent data. A store that states its facts plainly on its own domain is doing exactly what both systems are now looking for.
Where we fit
Most stores we look at believe their facts are on the site, and they are, for a human. The gap only shows up when you check the way a machine checks: which prices are rendered by script, which specs are trapped in images, which policies live in a PDF, where the feed and the page disagree. Our free AI Shopping Visibility audit does that check for you. You send a store URL and an email, and within three business days you get a prioritised report of what an AI can and cannot read on your product and policy pages, with no sales call attached. If the report shows work worth doing, our AI Shopping Visibility service fixes it across your whole catalogue, typically in about two weeks, and verifies it against the checks AI agents actually use. Retaining us means that when an AI comes to your site to confirm a price, the price is there to confirm.
Sources
- Ahrefs: How ChatGPT Does Query Fan-Out (15 Million Queries Studied) (12 October 2026; the fan-out, site: search, official-source and citation-share findings and their caveats)
- Google: More ways to connect with the web from Search (15 October 2026; the test of a top-of-answer section for official sites, inline links and Preferred Sources)
- Search Engine Land: Google updates AI Mode and AI Overview with more links, context, carousels and information on creators (15 October 2026; coverage of the changes and the lack of click data)
- PPC Land: Google pushes links higher in AI answers as Preferred Sources use rises 7x (analysis of what is a test and what is rolling out)
- JTS Tech: What ChatGPT actually reads from your site (19 August 2026; why most answers never open a page)
- JTS Tech: AI agents read your site through the same structure a screen reader uses (26 September 2026; why a fact a person can see may be invisible to an agent)
