How to Rank in ChatGPT: What the Citation Data Says

There is no ChatGPT leaderboard. When people ask how to rank in ChatGPT, they are really asking how to become the source an answer cites, or the business it names when a buyer asks what to choose. Kevin Indig analysed 3.7 million citations and found that only 2.37% of cited URLs appeared in all three of ChatGPT, Perplexity and Google AI Overviews. Three engines, three nearly separate contests. This guide covers the organic contest; the paid route into those same answers lives in our ChatGPT ads guide.

Precision matters here because the stakes stopped being niche: ChatGPT reached 900 million weekly active users, per OpenAI's February 2026 announcement, yet most advice about visibility inside it is folklore, recycled tips with no source in sight. Every claim below carries its study, and where the evidence is thin or vendor-shaped, we say so.

How does ChatGPT choose what to cite?

Mainly through live search, not model memory. Ahrefs studied 1.4 million prompts in April 2026 and traced 88.46% of ChatGPT citations back to its search retrieval layer. A page cannot be quoted until it has been fetched, which makes retrieval the gate every other tactic queues behind.

What happens after the fetch is where hopes die. In analysis Search Engine Land published in March 2026, Indig found roughly 30 domains collecting 67% of citations within a given topic, while 85% of the pages retrieved were never cited at all. The funnel narrows twice, once at retrieval and again at the moment of quotation, and that concentration carries a strategic read. If around 30 domains split two-thirds of a topic's citations, being averagely visible across ten topics is worth less than being one of the trusted few in one. Pick the topic where you can plausibly join that small club, then go deep on it. And treat the two gates as separate jobs: getting fetched is a technical visibility problem, covered next, while getting quoted is a writing problem, and the extraction data further down shows exactly what quoted passages look like.

	Google rankings	ChatGPT citations
What you optimise	Keywords and pages	Prompts and quotable passages
What you win	A position on one results page	A citation or a named recommendation
Carries across engines	One Google to worry about	2.37% of cited URLs shared across three engines
How you track it	Rank trackers	Manual buyer prompts, logged per engine

Can ChatGPT even see your site?

Open your robots.txt before touching anything else. OpenAI's crawler documentation draws a line many site owners miss: OAI-SearchBot governs whether pages can appear in ChatGPT search, while GPTBot governs model training only. Blocking GPTBot does not pull you out of search results.

The confusion dates to 2023, when a wave of sites blocked GPTBot to keep their content out of training runs. Reasonable decision, and it carries no search penalty. Blocking OAI-SearchBot is the act that takes you off the shelf. On mechanics, OpenAI states that it operates its own crawler for search and shares search queries with third-party providers, Bing among them, and we will not speculate past that on how the index behaves.

Slugs deserve an afternoon while you are in there. Across the Ahrefs dataset, URLs with natural-language slugs were cited at 89.78%, against 81.11% for opaque strings. A modest gap, but one of the cheapest levers on this list: name the page in plain words, in the address itself.

How do you get ChatGPT to recommend your business?

By earning mentions in the places the model already trusts, not only on your own site. Profound's February 2026 breakdown of citation origins put Wikipedia at 5% of all citations and Reddit at 3%, and 99% of those Reddit citations pointed at individual threads rather than subreddit homepages.

About Reddit, because you have probably seen a bigger number. The much-quoted 87% surge in Reddit citations was real, and it belongs to mid-2025, when Profound's citation tracking, reported by Search Engine Land that August, briefly had Reddit above 10% of all citations. By February 2026 the share had settled near 3%.

Quote the spike as history, the 3% as the present, and never blend the two.

Notice what that 99% figure implies. Nobody cites a brand's profile page; the engines cite the one thread where somebody solved the exact problem. A single genuinely useful answer in the right thread outweighs any volume of self-promotional posting.

Review platforms carry weight here too. G2 claims its pages are the most-cited B2B software source in AI answers, which is a vendor grading itself, so file it under claims rather than findings. The direction still checks out: when a buyer asks a model what to pick, structured third-party opinion is what it reaches for.

Practitioner evidence lines up as well. Ethan Smith of the growth agency Graphite told Lenny's Podcast that ChatGPT-sourced visitors convert at roughly six times the rate of Google visitors, and named three tactics working for his clients: dedicated landing pages, YouTube, and genuinely helpful Reddit comments. Those are one practitioner's numbers, without a public methodology, so read them as signal rather than benchmark.

The model recommends whoever the rest of the internet already recommends.

What pages that rank in ChatGPT have in common

They put the answer at the top. Indig's follow-up study of 1.2 million AI answers and 18,000 citations, published through Search Engine Land in February 2026, found 44.2% of citations sitting in the first 30% of a page. Content that saves its conclusion for the end mostly goes unquoted.

Where cited passages sit on the page

First third of the page

44.2%

Middle third

31.1%

Final third

24.7%

In the same Indig dataset, cited passages share a recognisable texture beyond position. They read like definitions or question-and-answer pairs, they run dense with proper nouns, around 20.6% of their words, and they sit at a simpler reading grade than typical prose. Length shows a sweet spot as well, with pages of 5,000 to 10,000 characters cited most.

None of this is new physics. The academic paper that coined generative engine optimization, presented by Aggarwal and colleagues at KDD 2024, measured visibility gains of up to 40% for the strongest of the content adjustments it tested. The shape you are reading right now, a question heading followed by a direct answer of 40 to 60 words, is that finding put to work, and we apply it to every post on this site.

Does schema markup help you rank in ChatGPT?

The evidence splits by engine, so the claim needs splitting too. Microsoft's Fabrice Canel confirmed at SMX Munich 2025 that structured data helps the language models behind Bing and Copilot. Ahrefs' 1.4 million-prompt study, meanwhile, surfaced no schema effect on ChatGPT citation at all.

Both results can be true at once.

Our practice follows that split, stated plainly rather than conveniently. Every post here ships FAQ schema, because it is cheap, it is evidenced for Bing's models, and structured pages tend to be well-organised pages anyway. What we never do is promise that markup earns ChatGPT citations, because the strongest dataset available says it does not. Why the engines differ is not something either source explains, and we are not going to guess. If someone sells you schema as a ChatGPT ranking lever, they are selling past the evidence.

Measuring your ChatGPT visibility, engine by engine

Use prompts a real buyer would actually type, and log every engine separately. That 2.37% overlap figure is why a single blended visibility score is unsound: each engine cites an almost entirely different set of URLs. Run your buying prompts inside ChatGPT itself, log every brand and link that comes back, and repeat monthly.

A tooling market has formed around this job at remarkable speed. G2's answer engine optimization software category opened with 7 products and passed 150 within ten months, as of January 30, 2026. Growth like that proves demand; it does not make the scores comparable, so keep manual prompt logs as your ground truth. Repeatable monthly watching is also exactly the kind of work we hand to AI marketing agents.

The most strategic finding in the measurement research is about brands, not domains. A Semrush study with Indig covering 50,000 brands, published on July 20, 2026, found that only 15.2% of categories have a settled winner the engines consistently name, and that branded search volume was the only statistically significant differentiator. Domain metrics did not separate leaders from the pack. Brand demand did, which means most categories remain open.

The part that compounds

Zoom out and the pattern is bigger than one engine. SparkToro and Similarweb measured 68.01% of US Google searches ending without a click between January and April 2026. Clicks are giving way to answers everywhere, and the citation is the unit that survives the shift.

Which brings back the paid comparison from the top, walked through in the step-by-step setup walkthrough. An ad stops the moment its budget pauses; a page the engines trust keeps earning citations while you sleep. Run both if the budget allows. But learn how to rank in ChatGPT organically regardless: one compounds, the other meters. Teaching machines to read your business well is most of our build work, and the first step costs a conversation, not a contract.