--- title: "How brands get cited in ChatGPT responses" excerpt: "ChatGPT cites a brand when it runs a web search and picks a page about that brand as a source. Most cited links are earned media, so coverage on other sites matters as much as your own. Your own pages win when OAI-SearchBot can crawl them, the answer sits near the top, and every claim links to a source." date: 2026-10-27 author: GeoVector readTime: 11 min read category: guide tags: [] --- # How brands get cited in ChatGPT responses *ChatGPT cites a brand when it runs a web search, retrieves the brand's page or a third-party page about it, and picks that page as a source. Most cited links are earned media, so coverage on other sites matters as much as your own pages. Your pages win when they let OAI-SearchBot crawl them, answer the question near the top, and back each claim with a source.* A ChatGPT citation is a linked source shown beside an answer. A mention is your brand's name in the answer text, with or without a link. You can earn one without the other, so track them as two numbers. GeoVector, a GEO (generative engine optimization) platform, tracks both for your brand across 6 AI assistants every week, which is how you see whether the steps below move anything. This guide covers how ChatGPT picks its sources, what makes a page citable, and why most citations come from sites you do not own. ## When ChatGPT cites sources at all ChatGPT only shows citations when it searches the web, and many answers never trigger a search. Ahrefs looked at 66.6 million free, logged-out ChatGPT responses from August and September 2026. Only 14.4% of them ran a web search [^fn-2]. That splits the work in two. On prompts that trigger a search, you compete for a citation. On the rest, ChatGPT answers from what the model already knows, so your brand can only appear as an unlinked mention. Prompts that ask for current facts, prices, comparisons, or "best" lists are the ones most likely to need fresh sources. Those are also the Research and Decision prompts where a buyer is closest to choosing. If you sell software, a prompt like "best expense tools for a 50-person company" is far more likely to cite sources than "what is an expense report". ## How ChatGPT picks which pages to cite ChatGPT rewrites the user's question into search queries, retrieves candidate pages, and then cites only a subset of them. A study of ChatGPT Web from GPT-5.2 to GPT-5.5 traced that process step by step [^fn-3]. Its central finding is that being retrieved and being cited are two different outcomes. Three results from that study matter for your planning: - **Consistency wins.** Across repeated runs, pages retrieved more often were also more likely to be cited [^fn-3]. One lucky appearance is worth little. - **Thinking modes search more.** The Instant settings mostly ran one retrieval round, while GPT-5.4 Thinking often ran several [^fn-3]. A page has more chances to surface in those modes, and more competition. - **Google rankings are a weak proxy.** Google Search and the OpenAI Responses API together recovered only about one quarter of the URLs ChatGPT Web actually cited [^fn-3]. That last point is the one most teams miss. A page that ranks first on Google can still go uncited in ChatGPT, and the reverse also happens. The authors' advice is to measure citation visibility directly and to re-check after each model change [^fn-3]. ### Which kinds of sources are gaining ChatGPT has been moving toward original sources and away from sites that summarize them. In Ahrefs' tracking of U.S. answers from March to September 2026, PubMed climbed from 64th to 4th among the most-cited domains, while WebMD fell from 18th to 79th [^fn-2]. Government, research, and regulator sites rose, along with hands-on product testers. Health and finance explainers lost ground [^fn-2]. Community sites are not a safe bet either. Reddit was cited in 19.1% of U.S. answers in March and 2.3% in September in the same dataset [^fn-2]. Wikipedia fell from 14.1% to 2.9% [^fn-2]. The lesson for a brand is direct. If your blog mostly rewrites other people's research, ChatGPT can cite the original instead. Publish what only you can publish: your own data, product documentation, pricing, and tested results. ## Make sure ChatGPT can reach your pages ChatGPT can only cite pages its search crawler is allowed to read. OpenAI runs OAI-SearchBot for search and GPTBot for model training, and the two are controlled separately in robots.txt [^fn-1]. OpenAI states that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers [^fn-1]. So you can block training and still be citable. Allow OAI-SearchBot, and disallow GPTBot only if you have a reason to. OpenAI says robots.txt changes take about 24 hours to reach its search systems [^fn-1]. Check your CDN and firewall as well as robots.txt. Bot protection rules often block crawlers that robots.txt allows, and nothing in your analytics will flag it. Test with your server logs, not just the robots.txt file. Many practitioners also recommend verifying your site in Bing Webmaster Tools, on the view that ChatGPT's search layer draws on a Bing-fed index [^fn-9]. OpenAI does not name its search providers in its crawler documentation [^fn-1]. Treat Bing indexing as a cheap precaution, not a guarantee. ## Write pages that are easy to cite ChatGPT cites passages, not whole pages, and it favors passages near the top. Kevin Indig's analysis of 18,012 verified ChatGPT citations found that 44.2% came from the first 30% of a page [^fn-4]. Content buried deep was roughly 2.5 times less likely to be cited [^fn-4]. Put the answer first. Open each section with two or three sentences that answer its heading on their own, then add detail, examples, and caveats. If your product page opens with a brand story, move a plain description of what the product does to the top. MaxAEO describes a quick check it calls the sentence extraction test [^fn-8]. Pick five paragraphs from a page and ask whether each one could stand alone as a defensible answer. Paragraphs that lean on vague claims like "best-in-class" fail it [^fn-8]. ### Back every claim with a source Specific, sourced claims give ChatGPT something to verify and quote. A line like "cuts invoice processing time" is hard to cite. A line that gives the before and after numbers, the sample, and the date, with a link to the write-up, is easy to cite. Use clear headings, short paragraphs, and tables for comparisons. Add Organization, Article, Product, or FAQ schema markup where it fits, so crawlers can tell who published the page and what it describes [^fn-9]. Schema will not rescue a weak page, but it removes doubt about a good one. ## Earn citations on other sites Most of what ChatGPT cites is not on brand websites. Muck Rack's Generative Pulse research found that 84% of citations in AI answers come from earned media [^fn-5]. Its May 2026 study covered more than 25 million links across ChatGPT, Claude, and Gemini, and paid or advertorial content made up just 0.3% of citations [^fn-6]. That 84% covers three engines, not ChatGPT alone. Still, the direction is clear. If you invest only in your own site, you are competing for a small slice of the available citations. Mentions on other sites also track AI visibility closely. Ahrefs analyzed 75,000 brands and found branded web mentions correlate with AI visibility at 0.66 to 0.71 across ChatGPT, Google AI Mode, and Google AI Overviews [^fn-7]. Classic authority metrics did much worse for ChatGPT: branded search volume correlated at 0.352 and Domain Rating at 0.266 [^fn-7]. Correlation is not proof of cause, and Ahrefs says so [^fn-7]. But a brand that few sites discuss should not expect ChatGPT to discuss it either. Practical ways to earn that coverage: - **Publish original data** that journalists and analysts want to cite. - **Get into independent reviews** and comparison pages in your category. - **Offer expert commentary** to trade publications that cover your buyers. - **Keep third-party profiles accurate**, so every mention describes you the same way. Paid placements are a poor shortcut. At 0.3% of citations, advertorial content rarely reaches the answer [^fn-6]. ## Why your own pages still matter An owned-page citation still pays off, because it is linked to more recommendations. 180 Marketing studied 10,000 ChatGPT responses and found brands were recommended 44% more often when one of their own pages was cited as a source [^fn-10]. So run two tracks at once. Build earned coverage to widen the set of pages that mention you. Make your own pages citable so ChatGPT can use them as a source when it does search. ## Measure ChatGPT citations directly You cannot infer ChatGPT citations from Google rankings, as the proxy study showed [^fn-3]. You need to run the prompts your buyers ask and record what comes back. Track three numbers per engine: | Metric | What it tells you | Why it matters | | --- | --- | --- | | Citation share | How often your pages appear as linked sources | Shows whether your own content is being used | | Mention rate | How often your brand is named in answers | Covers answers with no web search | | Brand share | Your position-weighted share of the answer versus competitors | Shows whether you are winning the prompt | Measure on a fixed prompt set, on a fixed schedule, and engine by engine. ChatGPT changes how it searches between model versions [^fn-3], so one snapshot goes stale fast. Compare before and after each content change so you know which change moved the number. This is the job GeoVector does. It runs a prompt set generated for your category across ChatGPT, Gemini, Claude, Perplexity, Google AI Overview, and Google AI Mode each week. It then reports citations, mentions, and brand share against your competitors, mapped to the Discovery, Research, and Decision stages. You can see which prompts you are losing and which cited pages your competitors are winning with. ## A checklist for your next 30 days Start with the steps that cost the least and unblock the rest: 1. **Open the door.** Allow OAI-SearchBot in robots.txt, check your CDN rules, and confirm in server logs that it reaches you [^fn-1]. 2. **Pick your prompts.** List the Research and Decision questions buyers ask in your category, the ones most likely to trigger a search. 3. **Fix your top pages.** Move a direct answer into the first 30% of each key page and source every claim [^fn-4]. 4. **Replace summaries with originals.** Swap rewritten explainers for first-party data, documentation, and tested results [^fn-2]. 5. **Plan earned coverage.** Choose two or three independent sites your buyers read and give them a reason to cover you. 6. **Set a baseline.** Record citation share, mention rate, and brand share before the changes, then re-measure weekly. ## Frequently asked questions ### Can you pay OpenAI to get your brand cited in ChatGPT? There is no sign that paying works. Paid or advertorial content made up just 0.3% of citations in Muck Rack's May 2026 study of more than 25 million links [^fn-6]. Earned coverage and well-sourced owned pages are the routes that work. ### Why does ChatGPT cite review sites instead of my website? ChatGPT has shifted toward original sources and hands-on product testers, and away from pages that summarize others [^fn-2]. Independent reviews often contain tested, specific evidence that brand pages lack. Most citations across ChatGPT, Claude, and Gemini come from earned media [^fn-5]. ### Does blocking GPTBot stop my site from being cited? No. GPTBot controls training use, and OAI-SearchBot controls whether your pages appear in ChatGPT search answers [^fn-1]. You can disallow GPTBot and still allow OAI-SearchBot. ### Why does ChatGPT mention a competitor instead of my company? Usually because the competitor appears on more of the pages ChatGPT retrieves, and appears consistently across runs [^fn-3]. Branded web mentions correlate strongly with AI visibility [^fn-7]. Check which pages are cited for your target prompts, then earn coverage on those sites or publish a better primary source. ### How often should I check my ChatGPT citations? Weekly is a sensible default. Citation patterns shift between model versions and months, as the swings in Reddit and Wikipedia citations show [^fn-2][^fn-3]. A single snapshot can mislead you. ## Sources [^fn-1]: [developers.openai.com](https://developers.openai.com/api/docs/bots): OpenAI's crawler documentation for OAI-SearchBot and GPTBot, including the search opt-out rule and the 24-hour robots.txt delay. [^fn-2]: [searchenginejournal.com](https://www.searchenginejournal.com/chatgpt-most-cited-sites-primary-sources-ahrefs/592682/): Ahrefs data on ChatGPT's most-cited domains from March to September 2026, Reddit and Wikipedia citation shares, and the 14.4% web search rate. [^fn-3]: [arxiv.org](https://arxiv.org/abs/2610.15382): Understanding ChatGPT Citations, an observational study of retrieval and citation in ChatGPT Web from GPT-5.2 to GPT-5.5. [^fn-4]: [growth-memo.com](https://library.growth-memo.com/p/ai-citation-brand-citation): Kevin Indig's analysis of 18,012 ChatGPT citations and their position on the page. [^fn-5]: [generativepulse.ai](https://generativepulse.ai/): Muck Rack Generative Pulse research on the share of AI citations that come from earned media. [^fn-6]: [everything-pr.com](https://everything-pr.com/answer-engines-how-ai-decides-which-brands-get-cited): Summary of Muck Rack's May 2026 study of more than 25 million cited links, including the 0.3% paid share. [^fn-7]: [ahrefs.com](https://ahrefs.com/blog/ai-brand-visibility-correlations/): Ahrefs study of 75,000 brands correlating search and web factors with AI visibility. [^fn-8]: [maxaeo.ai](https://maxaeo.ai/blog/chatgpt-search-citations/): Citations versus mentions in ChatGPT search and the sentence extraction test. [^fn-9]: [seoalive.com](https://seoalive.com/blog/rank-in-chatgpt): Practitioner guide covering Bing indexation and [Schema.org](https://Schema.org) structured data for ChatGPT. [^fn-10]: [180marketing.com](https://www.180marketing.com/chatgpt-cited-sources-brand-recommendations/): 180 Marketing study of 10,000 ChatGPT responses on citations and brand recommendations.