I get sent screenshots by clients almost weekly now. Someone’s asked ChatGPT to recommend a boiler installer, an electrician, or a renewable energy company in their area, and their competitor’s name comes back instead of theirs. It’s a genuinely uncomfortable moment for a business owner to see, and it’s exactly why this has become one of the most common questions I get asked on discovery calls.
I run Essheo, a search marketing agency working across the UK and US, and getting brands cited inside ChatGPT and other AI platforms is now a core part of what we do for clients.
What I want to do here is walk you through, honestly and specifically, how ChatGPT actually decides who to mention, because the mechanics are far more knowable than most businesses realise, and far more different from classic Google SEO than most people expect.
How Does ChatGPT Actually Decide Who to Cite?
ChatGPT operates on two entirely separate mechanisms, and understanding which one is active matters enormously. In its default, parametric mode, it answers purely from training data with no live retrieval and no citations at all.
In search or browsing mode, it retrieves live web pages and cites a small, specific number of sources, typically somewhere between 3 and 8 per answer depending on the query.
When browsing mode triggers, the model rewrites your question into one or more search queries, sends those to its retrieval index, opens a handful of results, extracts relevant passages, and synthesises an answer that names sources it can confidently attribute specific claims to.
A tactical guide from Tygart Media, analysing this process closely, found that ChatGPT Search actually cites only around 15% of the pages it retrieves. The other 85% get pulled into context, evaluated, and silently discarded with no visibility and no referral.
Why Does ChatGPT’s Retrieval Layer Run on Bing, Not Google?
This is the single most important technical fact I share with clients, and it consistently surprises even experienced in-house SEO teams. ChatGPT Search’s real-time retrieval layer is powered by Bing’s index, not Google’s.
That means if your site has never been properly submitted to Bing Webmaster Tools, has unresolved Bing-specific crawl errors, or simply isn’t well indexed by Bing, you may be functionally invisible to ChatGPT regardless of how well you rank on Google.
I’d genuinely recommend checking your top 10 ranking share for your core seed queries on Bing specifically, not Google, because that benchmark tells you far more about your ChatGPT visibility than any Google Search Console report ever will. This is one of the most overlooked, easily fixed gaps I come across when auditing new client accounts.
What Is Query Fan-Out and Why Does It Matter?
ChatGPT doesn’t run one broad search per question any more. It decomposes your prompt into several targeted sub-queries, a process researchers call query fan-out, and increasingly directs a meaningful share of those sub-queries at specific domains it considers authoritative for that topic, using the site-specific search operator rather than the open web.
Analysis by monitoring platform Promptwatch found that between August 7 and August 8 2026, the share of ChatGPT’s sub-searches using this targeted, domain-specific operator jumped from 0.37% to 16.8% of all queries in a single day, roughly 46 times higher, while the average number of searches per answer nearly doubled from 1.08 to 1.83.
The practical consequence was stark. Reddit’s share of ChatGPT citations collapsed 86.4%, from an average of 3.83% between 18 July and 7 August down to just 0.52% by mid-August. No policy change, no ban, just a shift in which domains the model decided were authoritative enough to query directly.
I mention this because it proves a point I make constantly to clients: citation share in AI platforms isn’t fixed or guaranteed. It can shift dramatically overnight based on model updates you have no control over, which is exactly why ongoing monitoring matters more than a one-off optimisation project.
What Signals Actually Increase Your Chances of Being Cited?
Once you understand that ChatGPT is evaluating extractability and authority rather than simply matching keywords, the practical tactics become much clearer. Research from SealGlobal, based on auditing more than 4,200 ChatGPT answers across five industries, found that the model doesn’t necessarily cite the objectively best content.
It cites the most extractable content from the most recognisable entity, on the most trusted source it can find inside its search layer.
Does Your Content Need to Answer the Question Immediately?
Yes, and this is one of the clearest, most actionable findings across the research I’ve reviewed. The model extracts the answer itself, not the surrounding paragraph, so putting a direct, complete answer within the first 200 words of a page significantly increases the chance it gets pulled and cited.
SparkToro’s separate research on LLM citation patterns found that 44.2% of citations are pulled from the first 30% of a page’s content, which tells the same story from a different angle. Bury your best answer under three paragraphs of preamble and you’re actively reducing your citation odds.
FAQ schema built around the actual questions your customers ask, phrased naturally with concise 50 to 80 word answers, is one of the highest-value structural tactics I’d recommend implementing. Pull those questions directly from tools like Semrush’s “People Also Ask” data or AlsoAsked rather than guessing, because matching real search intent language matters more than matching what you assume customers ask.
How Much Does Third-Party Corroboration Matter?
More than most businesses expect. ChatGPT’s citation-selection framework weighs domain authority, brand mentions already present in training data, and earned-media placement as some of the strongest structural signals available. Being cited across 3 to 8 listicles, comparison pages, or trade-press articles in your category is a genuinely stronger citation signal than picking up 50 generic backlinks.
This is backed up by hard data on what ChatGPT actually cites for shopping and recommendation queries. Analysis by Azoma of millions of shopping agent citations in Q2 2026 found ChatGPT’s source mix breaks down as 41% earned media, 37% retailer listings, 19% user-generated content, and just 3% brand.com pages.
In other words, if your GEO strategy leans entirely on your own website content, you’re addressing barely a third of what ChatGPT actually draws from, and paid ad placements sit entirely separate from the organic recommendation itself, meaning you cannot simply buy your way into being named.
Do Freshness and Technical Crawlability Genuinely Affect Citations?
Very much so. Visible lastUpdated dates and article:modified_time meta tags, along with current-year framing in titles and headings, materially boost citation rate compared with evergreen but undated content.
I’d recommend a quarterly content refresh cycle at minimum for anything you want to keep earning citations, because stale content quietly loses its place over time as fresher, better-dated competitors take over.
On the technical side, four things matter specifically for ChatGPT’s crawler: an llms.txt file, avoiding overly aggressive bot blocks in robots.txt that might be catching GPTBot or OAI-SearchBot, server-rendered HTML for your primary content rather than content that only loads via client-side JavaScript, and a clean, current sitemap.
I’ve seen genuinely strong content get zero AI citation simply because a robots.txt file was inadvertently blocking the crawlers that would have found it. It’s one of the first things we check in any technical audit at Essheo.
What Should Businesses Actually Do Differently Starting Now?
Bringing this all together, I think there’s a clear, ordered set of priorities that I’d walk any client through, and it starts with the technical foundations before moving into content and authority building.
Where Should a Business Start?
Get properly indexed and crawlable on Bing first, since that’s the retrieval layer ChatGPT Search actually runs on, not Google. Then confirm GPTBot and OAI-SearchBot aren’t being blocked in your robots.txt, and put an llms.txt file in place.
These are unglamorous technical fixes, but I’d argue they’re the single biggest quick win most businesses are missing entirely, because they’ve never been told ChatGPT’s retrieval mechanics differ from Google’s.
From there, audit your existing top content and restructure it so the direct answer sits in the opening lines, add FAQPage and Article schema using real customer questions, and build out a genuine topical cluster of 8 to 15 related, internally linked pages, since the model appears to use internal-link density as a proxy for topical authority.
A single long page rarely outperforms a properly linked cluster covering the topic from multiple angles.
How Should a Business Track Whether This Is Working?
Run your own category prompts through ChatGPT regularly and record exactly how your brand is mentioned, or whether it’s mentioned at all, rather than checking sporadically by hand.
Given how quickly citation patterns can shift, as the Reddit example shows, this needs to be an ongoing monitoring habit, not a single audit you file away and forget. This is exactly the kind of continuous tracking we build into client reporting at Essheo, because a citation you earned in July can quietly disappear by September if the model’s retrieval behaviour changes.
I’d also recommend building genuine third-party presence deliberately rather than hoping it happens organically. Earned media coverage, trade press mentions, comparison articles, and a presence on platforms with strong community trust signals all feed into the corroboration ChatGPT is actively looking for.
This is precisely the search everywhere approach we take at Essheo, treating your own site, third-party publications, forums, and video content as one connected system rather than separate, disconnected projects.
Ready to Find Out How ChatGPT Currently Talks About Your Business?
Most business owners I speak to have never actually checked what ChatGPT says when a potential customer asks about their category. That’s an uncomfortable blind spot given that a competitor could already be occupying the exact recommendation slot your business should hold, and given how quickly these citation patterns can shift without warning.
At Essheo, we combine deep technical SEO experience with a genuine understanding of how ChatGPT, Gemini and Perplexity actually source their answers. Every practitioner on our team carries 8 plus years of experience competing in hard, genuinely competitive sectors against household name rivals, and that same discipline, understanding exactly what earns trust and gets cited, is what we bring to AI search visibility work.
Our clients have generated over £45 million in combined revenue over the last two years through the search marketing systems we’ve built, and we treat ChatGPT visibility as one part of a wider, connected strategy across Google, LLMs, YouTube and social platforms, not a bolt-on afterthought.
We’ll start with an honest audit of exactly how ChatGPT currently represents your business and your competitors, then build a tailored roadmap to close that gap, covering the Bing indexing, technical crawlability, content structure and third-party authority building that genuinely move the needle.
Book a strategy call with me, and let’s find out exactly what ChatGPT is telling your customers about you right now, and what it would take to change that story.