ChatGPT SEO: how to get cited

How does ChatGPT choose the brands and sources it cites?

ChatGPT answers from two pools. The first is the knowledge acquired while the model was trained. Frozen and without links, it explains why a well-known brand gets cited without any search. The second is web search, triggered when the question calls for it: ChatGPT then reads pages and can display them as sources.

ChatGPT SEO is a branch of GEO, optimization for answer engines, presented in our GEO guide.

OpenAI documents its crawlers on a dedicated page (OpenAI, Overview of OpenAI Crawlers). Three of them directly affect ChatGPT SEO:

Crawler	Role stated by OpenAI	Respects robots.txt	Effect of blocking it
OAI-SearchBot	Surface websites in ChatGPT’s search features	Yes	Your pages no longer come up as sources in ChatGPT search
GPTBot	Collect content that may be used to train the models	Yes	Your content is left out of future training datasets
ChatGPT-User	Actions triggered by a user in ChatGPT or a custom GPT	Not necessarily: OpenAI states that robots.txt may not apply	Limited: the visit is requested by a person

OpenAI states that each setting is independent. A site can allow OAI-SearchBot to appear in search and refuse GPTBot. For search, a robots.txt change takes about 24 hours to be taken into account. Other providers’ crawlers are covered on our page GPTBot and AI crawlers.

ChatGPT SEO: cited as a brand or as a source?

These are two different outcomes, reached through different levers. Mixing them up leads to the wrong diagnosis.

	Cited as a brand	Cited as a source
What you see	ChatGPT names the company in its answer	A URL from your site appears as a reference, with a link
Where it comes from	The model’s knowledge and third-party pages read during search	Real-time web search on accessible pages
Main lever	Reputation: press, reviews, forums, knowledge bases	Technical access and pages that are easy to extract
Indicator	Mention rate across a panel of prompts	URL citation rate, pages cited

The gap between the two has been measured. Semrush recorded the answers of four engines (ChatGPT, AI Overviews, AI Mode, Gemini) across 115 prompts and 14 countries. Result: 62% of citations come with no mention of the brand name (Semrush, Ghost Citations Study, June 9, 2026). Your page can therefore serve as a source without your name appearing.

How do you rank on ChatGPT?

ChatGPT SEO follows the order of dependencies: access first, content next, brand on an ongoing basis. Here are the five workstreams.

1. Allow OAI-SearchBot in robots.txt, then check that the firewall or CDN does not block it.
2. Serve the content in HTML: the main text must be present in the source code, without relying on a script run in the browser (see JavaScript SEO).
3. Write citable pages: a direct answer at the top, self-contained sections, sourced and dated figures.
4. Build the brand outside the site: press, reviews, comparison sites, communities.
5. Measure on a fixed panel of prompts, and read the server logs.

Open access without being caught out by the CDN layer

A minimal robots.txt that allows search and refuses training looks like this:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

The training choice is a matter of content policy, not SEO. Robots.txt is not always enough: the CDN can filter crawlers before the file is even read.

Since September 15, 2026, Cloudflare has applied new default settings to new domains. Training crawlers and agents are blocked on pages with ads; search crawlers remain allowed (Cloudflare, July 1, 2026). Check the actual setting of your account instead of assuming it.

Make pages easy to cite

A cited page is rarely cited in full: the engine picks up a passage. Each section must therefore make sense on its own, with its topic, its answer and its context.

The founding GEO study measures a visibility gain when statistics and cited sources are added. It measures a loss with keyword stuffing (Aggarwal et al., arXiv, 2023). Its setup only puts five sources in competition per query: keep the ranking of the levers, not the percentages.

• a 40 to 60 word answer under the main heading;
• subheadings that use the real questions users ask;
• tables for any comparison, lists for any enumeration;
• an identified author and a visible update date.

Be present where ChatGPT looks

Semrush analyzed 126 million prompts between January and April 2026. ChatGPT cites about 15 sources per answer and often relies on Reddit and Wikipedia (Semrush, June 26, 2026). Your site is only one source among others. A plan limited to your own pages therefore covers only part of the cases.

According to an Ahrefs study of 75,000 brands, brand mentions are the signal most correlated with presence in AI Overviews, ahead of backlinks (Ahrefs, May 26, 2025). The study covers Google, not ChatGPT: it shows a trend, not a rule specific to OpenAI. Our GEO optimization method details this work.

How do you check whether ChatGPT cites your site?

OpenAI provides no dashboard for publishers. Checking relies on three sources, to be cross-referenced.

Source	What it shows	Limit
Prompt panel	Brand mentions, cited URLs, competitors named	Answers vary from one session to the next: run each prompt several times
Server logs	Actual visits from OAI-SearchBot, GPTBot, ChatGPT-User and response codes	Requires access to the logs
Analytics	Visits coming from chatgpt.com, conversions	Only captures clicks, which are a minority

For each reading, record the date, time, country, the model if the interface shows it, and the number of repetitions. The full method is described on our page on how to measure AI visibility.

Also keep the audience context in mind. According to Sensor Tower, quoted by TechCrunch, ChatGPT’s app market share fell below 50% at the end of May 2026 (46.4%). The app still has more than 1.1 billion monthly users (TechCrunch, June 16, 2026). Gemini and Claude are growing: tracking ChatGPT alone gives a partial view.

Which practices should you avoid?

• Hidden text or instructions addressed to AI: temporary loopholes, risk of penalty and risk to your image.
• The llms.txt file sold as a citation lever: no mainstream engine documents using it. See our analysis of the llms.txt file.
• Promises of results: ChatGPT’s selection rules are not published and change with the models.
• Working in isolation from SEO: ChatGPT search reads web pages. A site that is poorly crawled or poorly indexed starts at a disadvantage. Our article on AI SEO connects the two.

Solid ChatGPT SEO comes down to three decisions. Open access to OAI-SearchBot. Publish pages that answer quickly and cite their sources. Make the brand exist where ChatGPT reads. The rest is measured, prompt by prompt.