How to get cited by ChatGPT: what it takes to rank in it

The short answer

ChatGPT does not rank pages; it cites them, and it cites a narrow set of sources that most guides do not tell you to build. In TIN’s measurement of 299 ChatGPT citations, 68% went to government bodies and associations, and none went to a social or video platform source. Among 238 AI and automation agencies, 173 were never cited by ChatGPT source. So: get named by the institutions and publications in your field, be the primary record for something specific, let OpenAI’s search crawler read you, and stop paying for schema and formatting as citation levers.

Contents

1. What ChatGPT cites today
2. Why ChatGPT is the hardest engine to appear in
3. How ChatGPT picks a source: retrieval first, then citation
4. What to do, in order of the evidence
5. What not to pay for
6. How to check whether ChatGPT cites you
7. What this guide cannot tell you

What ChatGPT cites today

TIN asked seven AI engines the same 60 questions in 10 countries and coded every cited site into one of 10 categories: 1,395 cited sites in all source. The questions span 15 everyday topics: health, money, travel, food, technology, career, education, home, car, shopping, law, fitness, pets, parenting and business source. The study was pre-registered, its later changes are published as dated amendments source, and the dataset is downloadable from the study page.

ChatGPT’s answer to those questions leans on public bodies. Half of its 299 citations went to government sites, and associations took the next largest share source.

The 68% headline is the study’s own figure for government and associations combined: the two rows sum to 67.6%, which the study gives as 68% source. ChatGPT’s most-cited sites make the pattern concrete: consumer.ftc.gov (8), content.naic.org (7), pmc.ncbi.nlm.nih.gov (7), weforum.org (7), consumerfinance.gov (6) source.

The citations are spread wide below that top five. ChatGPT cited 174 different sites, and its top 10 sites hold 20.7% of its citations source.

Company sites took 16.4% of ChatGPT’s citations source. Gemini sends 58.3% of its citations to a company’s own site, the highest of any engine source.

The study’s own advice follows from the same row: “For ChatGPT, be cited by the institutions in your field. With 68% of its citations going to public bodies and associations, a business more often reaches ChatGPT’s answers through the regulator, trade body or standards group that covers it than through its own blog” source.

Why it barely overlaps with other engines

The engines disagree about which sites to cite. In the same study, 1,091 of 1,395 cited sites (78%) appear in only one engine’s answers source. The closest pair, AI Overviews and AI Mode, share 8 of their 25 most-cited sites; ChatGPT shares at most 1 with any other engine source.

YouTube is the most-cited site in Google AI Overviews and AI Mode; ChatGPT cites no social or video platform source.

The position this guide takes: ChatGPT is its own channel. The study puts it plainly: “Do not optimise for “AI” as one channel. With 78% of cited sites unique to one engine, a source strategy that works for ChatGPT says little about Gemini or Google” source.

Why ChatGPT is the hardest engine to appear in

The first study measured consumer questions. The second measured a business population: 238 agencies that sell AI or automation work as a primary offering, counted against the Ahrefs AI index on 2026-09-26 source.

“173 of 238 agencies were never cited by ChatGPT” source. The same population fared better elsewhere: “161 of 238 agencies were cited at least once by an AI platform, but the median agency was cited 3 times” source. Being visible on some engine did not carry over: “96 of the 161 agencies cited somewhere had no ChatGPT citations at all” source.

The study’s advice: “Expect ChatGPT to be the last place you show up” source. If you are tracking your company across engines and the engine is blank while another engine names you, that matches what 96 of the 161 cited agencies in this study saw source.

How ChatGPT picks a source: retrieval first, then citation

A citation is the last step of a pipeline, and a page can fail at any step before it.

Fetched. OpenAI’s search crawler has to be allowed in, and it has to find the content in the HTML. Steps 1 and 2 under “What to do” cover both.

Retrieved. The engine works on passages, not whole pages: “An answer engine fetches your page, splits it into blocks, scores each block on its own, and shows the model a handful of them” source. The full record, with each study’s limits, is in TIN’s post on the retrieval record.

Cited. Retrieval is not citation. One controlled experiment ran 252,000 trials across six LLMs over 18 content factors, testing which of two sources placed in the model’s context the first citation marker pointed to source. Four factors were unanimous across all six models source. Topic match came in at an odds ratio of 221 to >10k, a recent timestamp against a 2019 one at 14.4 to >10k, and, on product queries, a stated price at 6.26 to >10k source. The fourth unanimous factor is the source’s position in the list the model sees, so of the four gatekeepers in this experiment, a page controls three: topic match, a recent date and a stated price.

One change in 2026 matters for who wins. On 8 August 2026, Promptwatch recorded ChatGPT Search’s use of the site: operator in its background searches jumping from 0.37% of fanout queries to 16.8%, inside a single day source. A site: search names the domain it looks in. TIN’s inference, not a finding: that favours the site that is the primary record for a topic. Reddit’s share of ChatGPT Search citations fell from an average 3.83% (18 July to 7 August 2026) to 0.52% through 17 August, a drop the vendor itself calls provisional source. TIN’s reading of that shift is in the Reddit post.

What to do, in order of the evidence

Each step below names what supports it and how strong that support is. Most of the evidence is correlation. The controlled evidence is narrow, and it is labelled where it applies.

1. Let OpenAI’s search crawler in

OpenAI’s crawler documentation lists four agents source. One of them, OAI-AdsBot, is “used to validate the safety of web pages submitted as ads on ChatGPT” source, so it matters only if you advertise there. The three below decide search and training, and they do different jobs.

Each setting is independent. OpenAI gives the example of a site that allows OAI-SearchBot to appear in search results while disallowing GPTBot, so its content is not used for training source. In robots.txt that looks like this:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

Blocking GPTBot is a training decision, not a citation decision. Blocking OAI-SearchBot is the one that removes you from ChatGPT’s search answers, though OpenAI says opted-out sites “can still appear as navigational links” source. OpenAI notes that “it can take ~24 hours from a site’s robots.txt update for our systems to adjust” source. Check your robots.txt, and check any firewall or bot-management rule in front of it, because a rule that blocks unknown bots blocks this one too.

Strength: OpenAI’s own documentation.

2. Put every fact in the HTML

Vercel and MERJ logged AI crawler traffic and reported: “None of the major AI crawlers currently render JavaScript. This includes: OpenAI (OAI-SearchBot, ChatGPT-User, GPTBot)” source. That finding is dated 17 December 2024 source. A price, a table or a case figure that only appears after a script runs does not exist for these crawlers. View the page source, not the rendered page, and confirm that the facts you want quoted are in it.

Strength: one log study, dated 17 December 2024.

3. Get named by the institutions and publications your buyers’ questions land on

This is the step the measurement points at hardest. On everyday questions, public bodies and associations take most of ChatGPT’s citations source. That study covers 60 everyday consumer questions and does not cover business-to-business questions source, so what follows is TIN’s inference beyond the data, not a measurement: for a business, the equivalent bodies are the regulator, trade body, standards group or professional association that covers your field, and being listed, quoted or published by them is the likeliest way into the sources the engine already reads.

Two correlation studies point the same way: Ahrefs’ study of 75,000 brands and TIN’s study of 238 AI and automation agencies. Ahrefs, across three engines including ChatGPT, found YouTube brand mentions the strongest correlate at about 0.737, and branded web mentions at 0.656 to 0.709 by platform, against about 0.194 for the number of pages on the site source. Mentions track AI visibility; the size of a site tracks it far less. In TIN’s agency study, across all AI platforms, “Ranking keywords and AI citations rise together (Spearman 0.92)” and “Cited agencies have a median Domain Rating of 41; agencies with no citations, 9” source. Those two are all-platform figures, not ChatGPT figures.

Strength: consistent direction across three measurements, no controlled test. TIN’s own study says it directly: “Nothing here shows that raising Domain Rating, rankings or case-study quality raises AI citations” source.

4. Be the primary record for something specific

A site: search names a domain source. The domain worth naming is the one that holds the original: the dataset, the documentation, the measured result, the page a scoped query would be right to land on. A summary of other people’s work gives a scoped search no reason to pick you over the source you summarised.

Original data is also what the institutions in step 3 cite. A trade body links to a survey; it does not link to a listicle.

TIN’s Recon studies are a worked example of the format: a question, a method fixed before the data, a downloadable dataset, and the limits printed on the page. That is the format, not proof that it works. TIN’s own AI visibility baseline is 0% source, and this guide claims nothing for TIN’s pages that TIN has not measured.

Strength: an argument from the mechanism and one vendor’s instrument reading. The site: shift is recorded from outside the engine, and OpenAI has not explained it source.

5. Match the question and keep the date current

Of the four citation gatekeepers unanimous across six models in the controlled experiment described above, a page controls three: topic match, a current date and, on product queries, a stated price source.

In practice: write the page for the question your buyer asks, in their words, and answer it near the top. Put a real, current date on the page and keep it true; an updated date on a page that did not change is a claim you cannot back. On product and service pages, state a price where you can.

Strength: the only controlled evidence in this list. It tested models choosing between passages placed in front of them, not the live search product, so it describes the mechanism, not your traffic.

6. Write passages that stand alone

The engine scores blocks, not pages source. A block that says “as we saw above” carries no meaning when it is lifted out alone. Put the claim, the entity name and the figure with its date in the same block.

On length, TIN’s retrieval post puts it carefully: “The range worth acting on is 150 to 300 words”, and it is “not a measured optimum” source. Use it as a working range, not a target to hit.

Strength: an agreed mechanism, with the passage length still unsettled.

7. Rank on search engines too, but know it is a different channel

Search authority and AI citations rise together in TIN’s agency study, across all platforms source. Ranking still matters. But a top-10 ranking does not translate into an AI citation one for one, even on Google’s own AI surface.

TIN’s Google study measured that overlap for Google AI Overviews: “In the US, 30.7% of top-10 organic domains are also cited in the AI Overview (152 of 495, across 69 searches)” and “63.0% of the domains US AI Overviews cite are not among that search’s top-10 organic domains (259 of 411)” source. Position helps: “US organic positions 1 and 2 are also cited in 44.9% and 46.4% of searches; position 7 in 21.1%” source.

Those figures are Google AI Overviews, not ChatGPT, from one crawl window on 2026-10-01 and 2026-10-02 source. TIN has not measured the same overlap for ChatGPT; how far a top-10 ranking carries into its citations is unmeasured.

Strength: measured for Google, not for ChatGPT.

What not to pay for

The evidence on three common tactics is below. Each line is from the primary study, read in TIN’s post on the controlled evidence.

Schema is not useless; it does other jobs on a site. It is not a citation lever on the evidence available, and the study that tested it says so. The same goes for llms.txt: publishing one costs little, and almost none of those files received traffic source.

If someone sells you any of these as the way to get cited by ChatGPT, ask them for the measurement.

How to check whether ChatGPT cites you

Most companies test the wrong thing. They type their own name into the engine, see a correct description, and conclude they are visible. That measures recognition, not recommendation.

One preprint tested 112 startups across 2,240 queries. Asked about a product by name, ChatGPT recognised it 99.4% of the time; asked the buyer’s question without the name, the products appeared 3.32% of the time source. TIN’s post calls it “One author, one preprint” source, so treat the gap as structural, not as a settled number. The useful number is the second one.

A check that measures something:

1. Write the buyer’s questions, without your name. The question a stranger asks before they know you exist: “who builds invoice automation for mid-size distributors”, not “what does [your company] do”.
2. Run each prompt several times. A survey of 45 studies relays Schulte et al. (2026), who observed daily source-level Jaccard scores of approximately 0.34 to 0.42 across four engines over 45 days and propose seven to eight repetitions per prompt source. One run is an anecdote.
3. Record the engine, the date and the exact prompt. A citation share with no instrument and no unit is not a number. “Our ChatGPT citation share is 2%” says nothing until you know whose panel, which prompts and what denominator source.
4. Note what is cited instead of you. The sites the engine names for your buyer’s question are your list for step 3 above: the institutions and publications to be named by.
5. Keep ChatGPT separate from every other engine. With 78% of cited sites unique to one engine source, a blended score hides the engine that matters to you.

What this guide cannot tell you

• Correlations, not interventions. The citation sources study says of its own findings: “These are associations in one snapshot, not causes” source. Mentions, rankings and Domain Rating track citations. None of the studies here shows that raising them raises citations.
• One snapshot of everyday questions. The 299 ChatGPT citations come from 60 everyday questions; business-to-business questions are not covered, and the study is one weekly run, a snapshot, not a trend source.
• One population, one index, one date. The agency study counts AI and automation agencies, from one index, Ahrefs, on 2026-09-26 source. Other industries may differ, and TIN has not measured them.
• Google is not ChatGPT. The ranking overlap figures are for Google AI Overviews only.

Read next

• Which websites AI engines cite, TIN’s study: the full source mix for all seven engines, with the dataset.
• Which agencies AI engines cite: the B2B population, and how citations track search authority.
• Google top 10 against AI Overviews: how far ranking on Google carries into Google’s AI answer.
• ChatGPT stopped citing Reddit: the site: shift of August 2026 and what it says about the primary record.
• How ChatGPT ranks passages: how passages get fetched, scored and dropped.
• Generative engine optimization: the controlled evidence: the full evidence table behind “what not to pay for”.

Get cited. If you want this measured for your own market and your buyers’ questions, tell us what you sell.

Sources

1. TIN, “AI citation sources” study, edition 2026-10-02. https://theinternetninja.com/recon/ai-citation-sources-2026/
2. TIN, “Agency AI citations” study, edition 2026-09-27, Ahrefs AI index 2026-09-26. https://theinternetninja.com/recon/agency-ai-citations-2026/
3. TIN, “Google top 10 vs AI Overviews” study, edition 2026-10-02. https://theinternetninja.com/recon/google-top10-vs-ai-overviews-2026/
4. TIN, “Generative engine optimization: what the controlled evidence shows”. https://theinternetninja.com/blog/generative-engine-optimization-what-the-controlled-evidence-shows-2026/
5. TIN, “ChatGPT ranks passages, not pages”. https://theinternetninja.com/blog/ai-search-optimization-what-the-retrieval-record-shows-2026/
6. TIN, “ChatGPT stopped citing Reddit”. https://theinternetninja.com/blog/chatgpt-stopped-citing-reddit-substitute-platforms-2026/
7. OpenAI, “Overview of OpenAI crawlers”, read 2026-10-16. https://developers.openai.com/api/docs/bots
8. Vishwakarma, Kumar and Jamidar, SIGIR 2026, controlled experiment on citation factors. https://arxiv.org/abs/2605.25517
9. Vercel with MERJ, “The rise of the AI crawler”, 17 December 2024. https://vercel.com/blog/the-rise-of-the-ai-crawler
10. Amit Prakash Sharma, “The Discovery Gap: How Product Hunt Startups Vanish in LLM Organic Discovery Queries”, preprint. https://arxiv.org/abs/2601.00912
11. Ahrefs, AI brand visibility correlations, 75,000 brands. https://ahrefs.com/blog/ai-brand-visibility-correlations
12. Ahrefs, schema and AI citations. https://ahrefs.com/blog/schema-ai-citations/
13. Ahrefs, llms.txt study. https://ahrefs.com/blog/llmstxt-study/
14. Martinez, survey of 45 studies, relaying Schulte et al. (2026) on run-to-run agreement. https://arxiv.org/abs/2607.14035
15. Promptwatch, ChatGPT site: operator in fanout queries, August 2026. https://promptwatch.com/data/chatgpt-site-operator-fanouts