How to get cited by ChatGPT: how it chooses its sources, and what to do
Getting cited by ChatGPT means being one of the pages it links, or one of the businesses it names, when someone asks it who to buy from. ChatGPT rewrites the question, searches through partner providers, ranks the results on factors OpenAI does not publish, and cites a few pages, about six per conversation in Profound's count. In our data it names the business asked about in 31.8% of answers and cites the business's own site in 0.3%.
- OpenAI: ChatGPT searches when a question would benefit from current information, through partner search providers, with the query rewritten first. Ranking factors are not published; placement is not guaranteed.
- One published requirement: allow OAI-SearchBot. Blocking GPTBot, the training crawler, does not affect search: OpenAI says each setting is independent.
- Citations are spread thin: Profound found the top ten domains hold only 12% of ChatGPT's citations. Ahrefs found only 12% of cited URLs in Google's top ten for the same prompt.
- In our data ChatGPT names the business in 31.8% of answers, recommends it in 21.6%, and cites its own site in 0.3%. It reads reddit.com, en.wikipedia.org, github.com most.
What a ChatGPT citation is
Responses that use web search may include citations. Select a citation to open its source. [...] Select Sources, when available, to view cited sources and other relevant links.
OpenAI Help Center, Searching the web with ChatGPT
A citation is an inline link in the answer, with a Sources panel that lists every page the answer drew on. For a business there are two ways in: the page you own is cited, or your name appears in the answer because a cited page recommends you. The second is far more common. In our 1,776 measured answers ChatGPT named the business asked about 31.8% of the time and cited that business's own site 0.3% of the time. OpenAI's help page adds the warning that applies to all of it: "Search results and citations can be incomplete, outdated, or incorrect."
A short history of ChatGPT search
| Date | What happened | Source |
|---|---|---|
| 25 Jul 2024 | SearchGPT prototype: "clear, in-line, named attribution and links". Sites "can be surfaced in search results even if they opt out of generative AI training". | OpenAI |
| 31 Oct 2024 | ChatGPT search launches for Plus and Team: "ChatGPT will choose to search the web based on what you ask." Partners include Associated Press, Axel Springer and Reuters; it "leverages third-party search providers". | OpenAI |
| 16 Dec 2024, 5 Feb 2025 | Search opens to all signed-in users, then to people who are not signed in. | OpenAI, update notes on the launch post |
| 28 Apr 2025 | Shopping results with images, prices and buy links; several citations per answer, each tied to the passage it supports. | Reported by Search Engine Journal |
| 29 Sep 2025 | Instant Checkout and the Agentic Commerce Protocol, built with Stripe: "Product results are organic and unsponsored, ranked purely on relevance to the user." | OpenAI (date from coverage) |
| 6 Oct 2025 | Apps in ChatGPT, built on MCP; Sam Altman cites 800 million weekly users at DevDay. | OpenAI; TechCrunch for the figure |
| 2026 | Ads for signed-in adults on the free and Go tiers in the US, "at the bottom of answers"; "Ads do not influence the answers ChatGPT gives you." | OpenAI (page undated) |
| 10 Aug 2026 | Restaurant reservations through OpenTable, Resy and Yelp inside answers. | OpenAI release notes |
| 1 Oct 2026 | "Try on" for clothing and accessories in shopping results; favourites. | OpenAI release notes |
| 7 Oct 2026 | GPT-6 with "Intelligent UI": answers that combine text, visuals and interactive elements. | OpenAI release notes |
Quotes from the OpenAI pages linked in the sources. Two rows carry dates from press coverage because the OpenAI page shows none.
When does ChatGPT search the web?
ChatGPT may search the web automatically when your question would benefit from current information.
OpenAI Help Center, Searching the web with ChatGPT
A buying question nearly always qualifies: who is good, what it costs, which to choose. Users can also start a search by hand, and web search is available on every plan, signed in or not. When ChatGPT does not search, it answers from what the model learned in training, which is older and less specific; for a business, the search path is the one that matters, because it is where today's reviews and listings come in. OpenAI publishes no figure for the share of questions that trigger a search.
How does ChatGPT choose its sources?
ChatGPT search sometimes partners with other search providers. When it does, ChatGPT search typically rewrites your query into one or more targeted queries that it sends those providers. [...] ChatGPT ranks search results using multiple factors intended to help users find relevant, reliable information. Placement is not guaranteed.
OpenAI Help Center
That is the whole of what OpenAI says about ranking. Three things follow. First, the question your buyer typed is not the query that gets searched: ChatGPT rewrites it, often into several. OpenAI's own example turns "what's the latest on the development of drugs that target CCR8 for cancer?" into "CCR8 immunotherapy drug development 2025" and then "CHS-114 conference 2025", and it "may send additional, more specific queries" after reading the first results. If memory is on, saved memories can shape the rewrite too. Second, the search providers are partners, and the privacy policies OpenAI points to are Microsoft's and Shopify's, not Google's. Third, the ranking is OpenAI's own and unpublished, so the only way to know what it picks is to ask it and read the answer.
The best outside measurement of how far that ranking sits from Google's is Ahrefs' study of 15,000 long-tail prompts in July 2025: only 12% of the URLs cited by ChatGPT, Gemini and Copilot appeared in Google's top ten for the same prompt, and for ChatGPT's in-text citations the figure was 8%. Ranking on Google helps you be found by the provider; it does not decide who ChatGPT quotes.
What OpenAI says a site must allow
To make a website eligible for inclusion, allow OAI-Searchbot to crawl the site and confirm that the website host or content delivery network allows traffic from OpenAI's published searchbot IP addresses.
OpenAI Help Center
OpenAI runs three crawlers with three jobs, and its documentation is explicit that they are independent: "a webmaster can allow OAI-SearchBot in order to appear in search results while disallowing GPTBot." A site that blocks all AI bots to keep its content out of training also removes itself from ChatGPT's answers. Changes to robots.txt take about 24 hours to register.
| User agent | What OpenAI says it does | Effect of blocking it |
|---|---|---|
| OAI-SearchBot | "used to surface websites in search results in ChatGPT's search features" | Your site is not shown in ChatGPT search answers |
| GPTBot | "used to crawl content that may be used in training our generative AI foundation models" | Your content is not used for training; search is unaffected |
| ChatGPT-User | visits a page when a user asks; "not used to determine whether content may appear in Search" | robots.txt rules may not apply, it is a user action |
From OpenAI's crawler documentation, read 24 October 2026. IP address lists: openai.com/searchbot.json, gptbot.json and chatgpt-user.json.
Which sites ChatGPT cites, in the open studies
Four studies with open methods, each measuring something different.
| Study | Sample | Finding |
|---|---|---|
| Profound, February 2026 | About 730,000 US conversations with at least one citation, October to December 2025 | Wikipedia in 18% of conversations and 5% of all citations; Reddit 13% and 3%; Reuters and NIH 4% each. The top ten domains together hold 12% of all citations. About six unique citations per conversation, rising through the quarter. |
| Semrush, November 2025 | 230,000 prompts, 14 July to 12 October 2025, over 100 million citations across ChatGPT, AI Mode and Perplexity | Reddit was cited in close to 60% of ChatGPT responses in early August and around 10% by mid September; Wikipedia fell from roughly 55% to under 20% in the same weeks. The mix changes without notice. |
| Ahrefs, July 2025 | 17 million cited URLs across six engines | Pages cited by ChatGPT averaged 958 days since publication, against 1,432 for Google's organic results. ChatGPT and Perplexity order in-text references from newest to oldest. |
| SE Ranking, November 2025 | 129,000 domains, 216,524 pages, 100,000 prompts, 20 niches | Domains with millions of Quora and Reddit mentions had roughly four times the citation likelihood. Pages updated within three months averaged 6.0 citations, outdated pages 3.6. Articles over 2,900 words averaged 5.1, under 800 words 3.2. An llms.txt file: negligible. |
Each study's own figures and sample. Profound counts conversations and citations, Semrush responses per week, Ahrefs page age, SE Ranking citations per domain.
What matters for a brand, as opposed to a page, is harder to pin down and the two Ahrefs studies disagree on the size of it. On the top 50 sites per engine in June 2025, branded web mentions correlated with ChatGPT visibility at 0.15 (Spearman), against 0.65 for AI Overviews. On 75,000 brands in December 2025, mentions across YouTube and the web were the strongest factors (about 0.74 and 0.66) and link metrics were "very weak". Different samples, one direction: being talked about on other sites predicts being named, and backlinks barely do.
What ChatGPT answers look like, measured
| Engine | Answers | Business named | Recommended | Own site cited |
|---|---|---|---|---|
| Perplexity | 1,777 | 32.2% | 20.1% | 0.6% |
| ChatGPT | 1,776 | 31.8% | 21.6% | 0.3% |
| Claude | 1,687 | 33.8% | 18.8% | 0.7% |
| Gemini | 1,543 | 40.7% | 26.1% | 0.5% |
| Google AI Overviews | 769 | 42.4% | 12.7% | 1.2% |
7,552 answers, 35 businesses, 24 August to 7 October 2026.
ChatGPT cites a business's own site less often than any other engine we measure, 0.3% of the time, while naming the business in 31.8% of answers. Its answers come from pages about you. The platforms it reads most in our data:
| Platform | ChatGPT answers citing it |
|---|---|
| reddit.com | 232 |
| en.wikipedia.org | 43 |
| github.com | 42 |
| checkatrade.com | 22 |
Platforms cited by ChatGPT across our answers, counted once per answer; listed only when answers about five or more businesses cite them.
Asked the same question on the same day, ChatGPT and another engine name the same business first 26 to 32.3% of the time, so a ChatGPT result says little about Gemini or Perplexity.
ChatGPT, local businesses and shopping
OpenAI's help page says ChatGPT may use an approximate location from the IP address, may rewrite "restaurants near me" into "top restaurants San Francisco", and, for restaurants, shows reservation availability from OpenTable, Resy or Yelp where the business is listed. For a local business that means two things: the question ChatGPT searches is a city-level query, so the pages that rank for "best X in [city]" are what it reads; and a listing on the reservation and review platforms it integrates with is a direct route into the answer.
For a shop the route is a product feed. OpenAI's feed specification asks for nine required fields (id, title, description, URL, brand, seller name, image, availability, price) "for a useful discovery feed", and warns that "eligibility does not guarantee display". Product results are "organic and unsponsored, ranked purely on relevance to the user", and Instant Checkout "items are not preferred". So the feed gets a product considered; the description and the pages about the product get it picked. Citeon's Shopify app measures that product by product.
How to get your business cited by ChatGPT
- Let OAI-SearchBot in. Check robots.txt and the CDN rules against OpenAI's searchbot IP list. This is the one rule OpenAI publishes, and sites that block every AI bot at once fail it. Keep GPTBot blocked if you want; it changes nothing here.
- Be on the pages ChatGPT reads. reddit.com, en.wikipedia.org, github.com in our data; Wikipedia, Reddit and the trade press in Profound's; Quora and Reddit mentions in SE Ranking's regression. Answers are built from these, not from your homepage.
- Answer the rewritten query. ChatGPT searches "best accountant Rotterdam freelancers", not your brand. Have a page that is the honest answer to that query, with the facts a model can quote: prices, scope, location, proof.
- Keep it fresh and complete. Cited pages were 958 days old on average against 1,432 for Google's results, and pages updated within three months drew nearly twice the citations. Long, specific pages beat short ones in SE Ranking's data.
- Keep the facts consistent. ChatGPT cross-reads sources. A price on your site that disagrees with your directory listing costs you the recommendation.
- Skip the folklore. No llms.txt effect was found. No markup is required. Backlinks correlate weakly; mentions correlate strongly.
- Measure it daily. Answers change from day to day, and the cited domains changed by half within six weeks in Semrush's data. One check is noise; a rolling average over four weeks is a trend.
Is this ChatGPT SEO?
People call it that, and the phrase mostly means something else: using ChatGPT to write titles and briefs for Google. This page is about the other direction, being chosen by ChatGPT. There is no ranking to climb; there is an answer, written for one person, that names a few businesses and often picks one. The work is being on the pages ChatGPT reads, being described accurately there, and having a site its search crawler can reach. It is one engine's slice of generative engine optimization.
How to measure it
Ask ChatGPT your buyers' questions on a schedule and record what came back: named, position, recommended, the sources cited, the competitors named. Citeon AI reads ChatGPT's own interface for this, daily, in your buyers' country, next to Gemini, Claude, Perplexity and Google AI Overviews, keeps every answer, and turns each losing question into a fix on your site. How the tool works, or start with the free audit.
Questions people ask
How do I get my website cited by ChatGPT?
OpenAI gives one requirement: allow OAI-SearchBot in robots.txt and let OpenAI's searchbot IP addresses through your host or CDN. After that it comes down to being a page ChatGPT would quote for the rewritten query, and being named on the pages it already reads.
How does ChatGPT choose its sources?
OpenAI's words: ChatGPT ranks search results using multiple factors intended to help users find relevant, reliable information, and placement is not guaranteed. It rewrites your question into targeted queries, sends them to partner search providers, and cites a few of the results.
Does ChatGPT use Google?
OpenAI's help page says ChatGPT search sometimes partners with other search providers; the privacy policies it points to are Microsoft's and Shopify's. It does not name Google. Ahrefs found only 12% of URLs cited by ChatGPT, Gemini and Copilot in Google's top ten for the same prompt.
Does blocking GPTBot stop ChatGPT citing me?
No. GPTBot is the training crawler and OAI-SearchBot the search crawler. OpenAI says each setting is independent and a site can allow OAI-SearchBot while disallowing GPTBot. Blocking OAI-SearchBot is what removes a site from ChatGPT search answers.
Which websites does ChatGPT cite most?
Profound's study of about 730,000 US conversations, October to December 2025: Wikipedia in 18% of conversations with citations, Reddit in 13%, Reuters and NIH in 4% each. The top ten domains together were only 12% of all citations, so most citations go to the long tail.
Does ChatGPT prefer fresh content?
Yes, measurably. Ahrefs' study of 17 million citations found pages cited by ChatGPT were 958 days old on average against 1,432 for Google's organic results, and ChatGPT orders its in-text references from newest to oldest.
Does llms.txt help you get cited by ChatGPT?
SE Ranking's regression on 129,000 domains found that an llms.txt file had a negligible effect on ChatGPT citation likelihood. Mentions on Quora and Reddit, content updated within three months, and longer articles all showed a measurable effect.
Why does ChatGPT recommend my competitor and not me?
Because of what it found, not what you wrote. In our data ChatGPT names the business asked about in 31.8% of answers, recommends it in 21.6%, and cites the business's own site in 0.3%. It builds the answer from third-party pages and names the businesses those pages recommend.
Are ChatGPT's product results paid?
No. OpenAI says product results are organic and unsponsored, ranked purely on relevance to the user, and that items with Instant Checkout are not preferred. Ads, where shown, sit at the bottom of an answer and do not influence it, in OpenAI's words.
How many people use ChatGPT?
OpenAI does not publish a running figure. Sam Altman said 800 million weekly users at DevDay on 6 October 2025, as reported by TechCrunch, which reported 900 million in February 2026. OpenAI publishes no share of questions that trigger a web search.
Sources
- OpenAI Help Center, Searching the web with ChatGPT. when it searches, query rewriting, providers, citations, the ranking sentence, OAI-SearchBot eligibility, location. Updated about 20 October 2026
- OpenAI, Overview of OpenAI crawlers (OAI-SearchBot, GPTBot, ChatGPT-User). "Each setting is independent of the others"; 24 hours for robots.txt changes
- OpenAI, SearchGPT prototype. 25 July 2024; "Sites can be surfaced in search results even if they opt out of generative AI training"
- OpenAI, Introducing ChatGPT search. 31 October 2024; updates of 16 December 2024 and 5 February 2025
- OpenAI, Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol. "Product results are organic and unsponsored, ranked purely on relevance to the user"
- OpenAI, Product feed specification. nine required fields; "Eligibility does not guarantee display"
- OpenAI, Introducing apps in ChatGPT. DevDay, 6 October 2025; built on MCP
- OpenAI, Our approach to advertising and expanding access. "Ads do not influence the answers ChatGPT gives you"
- OpenAI Help Center, ChatGPT release notes. 2026 entries: reservations, Try on, GPT-6 Intelligent UI
- Profound, How ChatGPT sources the web. Brandon Punturo, 3 February 2026; about 730,000 US conversations with a citation, October to December 2025
- Semrush, The most-cited domains in AI: a 3-month study. Luke Harsel, 10 November 2025; 230,000 prompts, 14 July to 12 October 2025
- Ahrefs, Only 12% of AI cited URLs rank in Google's top 10 for the original prompt. Louise Linehan, 11 August 2025; 15,000 prompts
- Ahrefs, AI assistants prefer to cite fresher content. Ryan Law, 28 July 2025; 17 million citations
- Ahrefs, Top brand visibility factors in ChatGPT, AI Mode and AI Overviews. Louise Linehan, 12 December 2025; 75,000 brands
- SE Ranking, How to optimize for ChatGPT. 24 November 2025; 129,000 domains, 216,524 pages
- TechCrunch, Sam Altman says ChatGPT has hit 800M weekly active users. 6 October 2025; the 900 million figure of 27 February 2026 is TechCrunch's too
- Citeon AI measurements, 24 August to 7 October 2026. 1,776 ChatGPT answers among 7,552