ChatGPT has started checking your facts before it recommends you. This is the clearest takeaway from Ahrefs' new study of 66.6 million ChatGPT responses and the 15 million fan-out queries behind them.1 The direction of travel is clear: when ChatGPT searches, it increasingly goes to whoever owns the information which means that your site now carries more weight than the sites that write about you.
This matters most for considered-purchase brands because, before ChatGPT recommends a brand, it runs the same checks a careful buyer does: is this the official source, what does it cost, what do customers say, and how does it compare? If your website can’t answer these types of questions when buyers search on AI platforms, ChatGPT will look for the answers somewhere else.
Here is our take on the data, and what we would do about it.
1.1 What the new fan-out data tells us
1.1.1 What query fan-out is
Query fan-out is the process an AI assistant uses when it needs current information. It breaks a user’s prompt into several background searches, runs them in parallel, and builds a single answer from what comes back. Take a prompt like “which business insurance broker should a 200-person firm use?”. A model might search the official sites of two or three brokers it already knows, look for information on fees, and search for reviews or comparisons. The end user only sees one answer after the response is generated but behind it sit several searchers, each looking for a different kind of evidence. One clear distinction matters: being retrieved is not the same as being cited by the model. This means the brands that get named/cited are the ones that answer the questions the model is actually asking.
1.1.2 Four findings that matter
- Most prompts never reach the web – Only 14.4% of the responses Ahrefs tracked triggered a search. For the rest, ChatGPT answered from what it already knew.
- When it does search, it goes to the source – Around three quarters of the most common site: searches targeted a brand’s own website, and the word “official” appeared in 13.1% of English fan-outs in September, up from 10.1% in August.
- It is evaluating rather than explaining – Fan-outs containing “price” rose by 47% in a month and “reviews” by roughly a third, while “meaning” and “definition” fell.
- Summaries are losing ground to originals – Reddit’s share of ChatGPT citations fell 88% in six months and Wikipedia’s by 80%. Regulators, research bodies and brand sites gained.
Ahrefs is clear that this is one model, one free tier and a two-month window, and that OpenAI changes how retrieval works often. The direction is still clear enough to act on.
1.2 Our take: fan-out is your buyer’s due diligence
ChatGPT adds the same words to its searches again and again: official, price, reviews, best and vs. Each one matches a question a buyer asks just before they decide. Is this the official source? What will it cost? What do other customers say? How does it compare with the alternatives?
These are confidence questions. When ChatGPT runs a fan-out search, it is looking for clear evidence that your brand is a safe recommendation, and it will only name you if it can find that evidence.
1.2.1 ChatGPT asks what your buyers hesitate over
We call the space between a buyer’s intent and their action the Confidence Gap™. It is where most considered-purchase brands lose people: in the messy middle, where buyers compare, hesitate and stall. Challenger’s research puts the share of complex deals that end in no decision at 60%. Each fan-out type maps to a confidence need. “Official” is about credibility. “Price” and “cost” are about value and clarity. “Reviews” are social proof. “Best” and “vs” are about reducing the risk of choosing wrong. When your brand answers these questions clearly, the model can recommend your brand with confidence. When it doesn’t, the model finds the answer elsewhere, and a competitor or third-party summary gets cited.
1.2.2 Why this hits considered-purchase brands harder
- Comparison is the default – Considered purchases always get compared. In Ahrefs’ test of 20,000 prompts, official sites earned 63.4% of citations for factual questions but only 46.4% for comparison questions. The answer to “A or B?” is mostly shaped by sources you don’t own.
- Pricing is often hidden – Many considered-purchase brands put price behind a “contact us” form. As price fan-outs rise, that gap gets filled by a competitor or a third party. If exact prices aren’t possible, publish ranges, starting points or the factors that drive cost.
- Regulated sectors face a higher bar – In insurance, finance, healthcare and legal, ChatGPT increasingly favours regulators and official bodies, with gov.uk among its 15 most site: searched domains. Your claims need to agree with official guidance, because they’ll sit next to it.
- Professional buyers probably trigger more searching – Ahrefs only studied free, logged-out sessions, but it notes that paid and reasoning models run several times more fan-outs. As B2B buyers often research on work accounts, treat the 14.4% search rate as a floor, not an average.
1.3 Two ways in: be known, or be the source
Ahrefs closes its study with a simple but useful idea. Either ChatGPT already knows your brand well enough not to look you up, or it finds you as the original source when it does search. Considered-purchase brands need to win on both fronts, and each calls for a different kind of investment.
1.3.1 Build AI brand visibility before the search starts
For most free prompts, the answer comes from what the model learned in training. Whether you appear depends on how widely, consistently and credibly your brand is discussed. Earned coverage in the right places now carries more weight. OpenAI’s licensed publisher partners roughly doubled their share of ChatGPT citations between March and July, from 7.5% to 15.8%, and still held 12.5% in late September.
1.3.2 Be the source when it does search
When ChatGPT runs a site: search on your domain, it is looking for facts only you can publish: pricing guidance, specifications, eligibility, service scope, support and contact details. Put them on your own website in plain, crawlable HTML, not in PDFs, images or widgets that only render in the browser. Also make it very clear which domain is yours – a site: search only works if the model knows your official domain, and smaller brands, or brands that share a name with others, are at risk of being confused. This means using your brand name and domain consistently across your listings, profiles and coverage such as LinkedIn, review sites, Google Business Profile, press releases and industry directories.
Original evidence matters just as much. Your own research, benchmark data and case studies with real numbers are things only your brand can publish, and Ahrefs found these are the pages holding up best, while pages that summarise other people’s work are losing citations. None of this replaces search. Fan-outs still run on search indexes, so ranking well for the themes ChatGPT searches is still how you get retrieved. Generative engine optimisation builds on SEO rather than replacing it.
1.4 What we would do now: Signals, Science, System
Knowing how fan-out works is only useful if it changes what you do next. We use the 7DOTS Confidence Framework to help considered-purchase brands turn customer interest and intent into action, and it maps closely onto the way ChatGPT now evaluates brands. Signals is about understanding buyer intent, behaviour and emerging opportunities. Science applies behavioural science to reduce uncertainty and increase trust. System creates joined-up digital experiences that consistently reinforce confidence at the right moment. Here is how each one applies to fan-out, and where we would start.
1.4.1 Signals: AI visibility tracking for your category
Start by finding out what ChatGPT actually checks in your market. Tools such as Ahrefs Brand Radar show the hidden searches it runs, which usually fall into a few types: official information, price, specifications, reviews and comparisons. The mix varies by sector, so let your own data guide what you fix first. You can also use tools you already have. In Google Search Console, searches like “site:yourbrand.com” with lots of views but almost no clicks are usually AI tools checking your site. Bing Webmaster Tools goes a step further: its AI Performance report shows how often Copilot and Bing’s AI answers cite your pages, along with a sample of the “grounding queries” used to find them, the closest thing to official fan-out data available today. Then ask ChatGPT your most important buyer questions a few times and see whether your brand comes up, using an AI visibility tool to track them at scale. Track how often it looks at your site and how often it names you, and link both to enquiries.
1.4.2 Science: how to optimise content for AI search
Make sure your most important pages answer the four questions ChatGPT is checking. Is this the official source? Is it clear what this costs? What do customers say? How does it compare? Fix the gaps in the order your data says they matter, and put the answer first, because clear claims and real numbers help buyers and models alike. Be honest in your comparisons, even where a competitor is the better fit, as self-promotional “best of” listicles are content ChatGPT is moving away from. Finally, strengthen your third-party proof on the review platforms, accreditations and publications that matter in your sector, including smaller specialist review sites, and monitor what they say.
1.4.3 System: make your website the source of truth
We often find that brands don’t have a single source of truth. Prices sit with sales, specifications with product, terms with legal, and over time the versions quietly drift apart. If ChatGPT finds three different answers, it has no clear one to give a buyer. Give each set of facts an owner and a date to review it, then make sure the same version appears on your website, your directory listings, your review profiles and in the press. Add structured data to the pages that hold these facts, and update the published date when something genuinely changes. After that, keep an eye on it. ChatGPT changes the way it searches every few months, and a quick monthly check will help you spot the next shift early.
1.5 What not to do
1.5.1 Don’t chase individual fan-out queries
When you first see the fan-out data, the natural reaction is to create a page for every query, but we’d advise against it, because most of these searches are one-offs that shift from person to person and churning out pages to match them can fall foul of Google’s rules on scaled content. You’ll get more from grouping the queries into clear themes and making sure your best pages cover them well.
1.5.2 Don’t over-read one study
Ahrefs found that ChatGPT ran about 1.6 fan-out queries each time it searched, while other studies have recorded far more, largely because they tested different models, account types and prompts. Rather than relying on any one headline figure, it’s worth building your own baseline by tracking the prompts that matter most in your category over several months. This is something we do for clients as part of our Generative Engine Optimisation (GEO) work, so they can see exactly how ChatGPT and other AI assistants behave for their own buyers, and spot changes as they happen.
1.5.3 Don’t write for the model instead of your buyer
Once teams start looking at fan-out data, it’s tempting to write for ChatGPT rather than for the people who actually buy from you, stuffing pages with words like “best” or “official” in the hope of being picked up. In our view, that gets it the wrong way round. The questions the model checks are the same ones your buyers ask, so the best approach is creative content that answers them clearly, sounds like your brand and earns trust, which serves the buyer and the model equally well.
1.6 AI search strategy: what this means for CMOs
For CMOs, AI visibility is now a confidence problem rather than a traffic problem. The first answer a buyer sees about your brand is increasingly assembled by ChatGPT from sources you only partly control. Here are five decisions that will put you back in charge of what that answer says.
- Fund both routes into the answer – Invest in brand and earned media so the model already knows your name, and in first-party facts and evidence so it finds you whenever it searches.
- Make pricing and proof a leadership call – Publishing price guidance and evidence is now a visibility decision, so agree it at board level rather than leaving it to marketing alone.
- Measure what actually drives pipeline – Track how often AI tools retrieve your pages and how often they name you, then connect both to enquiries rather than simply counting mentions.
- Review AI visibility every quarter – ChatGPT changes how it searches every few months, so build a regular review into your planning cycle rather than treating this as a one-off project.
- Start with your most valuable buyer questions – Identify what prospects ask before they buy, use an AI visibility tool to track what ChatGPT says about you, and fix the biggest gaps.
This is where our work sits. We help considered-purchase brands close the Confidence Gap™ by joining up SEO, Generative Engine Optimisation (GEO) and creative content, so your brand is easy to find, easy to understand and easy to recommend, for buyers and for AI models. If you want to know what ChatGPT says about your brand today and where the biggest gaps are, our team can show you.
1.7 The bottom line
ChatGPT is starting to behave like a careful buyer, checking the source, the price, the reviews and the alternatives before it commits to a recommendation. The details will change with each model update, but the principle won’t: the brands that win make it easy to feel confident choosing them.
That’s why the answer isn’t a set of quick tactics, but a system you can keep running. If you want to know where your brand stands today, our free 5Cs Snapshot gives you a one-page view of where confidence is won and lost across your digital journey, in a single 60-minute session with our team.
1.8 Frequently asked questions
1.8.1 What is query fan-out?
Query fan-out is the process an AI assistant such as ChatGPT uses to answer a prompt that needs current information. It splits the prompt into several background searches, runs them at the same time, and then combines everything it finds into a single, joined-up answer for the person asking.
1.8.2 Does ChatGPT always search the web before answering?
No. In Ahrefs’ study of free, logged-out sessions, only 14.4% of responses triggered a web search. Paid and reasoning models search more often and run more fan-outs per prompt.
1.8.3 How to check AI visibility?
Start by tracking the questions your buyers ask in an AI visibility tool such as Ahrefs Brand Radar, which shows how often your brand is mentioned and cited. You can also ask ChatGPT those questions yourself a few times, and check Google Search Console for “site:” searches on your own domain.
1.8.4 How to get cited in AI search?
Publish the facts only you can provide, such as pricing guidance, specifications and real case studies, in plain text on your own website. AI assistants increasingly go to the original source when they search, so clear, specific and consistent information gives them a good reason to cite you.
References:
1* Ahrefs Blog, https://ahrefs.com/blog/query-fan-out-chatgpt/, <Accessed 12 October 2026.
