How to get cited in ChatGPT search results

The 30-second answer: let OAI-SearchBot in, be in the index ChatGPT searches, and answer the question in the first paragraph

OpenAI’s publisher FAQ says any public website can appear in ChatGPT search, as long as it is not blocking OAI-SearchBot. That is the gate. Behind it, ChatGPT finds candidate pages through third-party search providers it does not name, and the evidence since launch points at Bing. So the levers are: allow the right bot, make sure Bing has your pages, and write each page so a model that quotes can quote it.

Five steps, in order of effect: allow OAI-SearchBot and ChatGPT-User in robots.txt and in your firewall; get your pages into Bing with Webmaster Tools, a sitemap, and IndexNow; make each important page answer one question as text; keep every copy of a fact consistent; and measure with the utm_source=chatgpt.com referrals OpenAI adds to its links.

How ChatGPT search finds and cites a page

OpenAI’s crawler documentation describes three agents with three jobs, and the distinction matters more here than for any other assistant:

• OAI-SearchBot is for search. OpenAI says it is used to surface websites in search results in ChatGPT’s search features, and recommends allowing it in robots.txt. A robots.txt change takes about 24 hours to be reflected. Its IP ranges are published at openai.com/searchbot.json.
• ChatGPT-User fetches a page when a person asks something that needs it. OpenAI says it is not used for crawling the web in an automatic fashion, that robots.txt rules may not apply because a user initiated the request, and that it is not used to determine whether content may appear in search.
• GPTBot is for training foundation models. Disallowing it indicates your content should not be used in training. It has no bearing on search.

The publisher FAQ adds two details site owners ask about. A page you have disallowed can still show up as a bare link and title if OpenAI finds the URL elsewhere and judges it relevant; it just cannot be summarised. To keep a page out entirely, use a noindex meta tag, and let the crawler fetch the page so it can read the tag. And ChatGPT adds utm_source=chatgpt.com to the links it shows, which is how you will measure all of this later.

Where Bing fits, as of October 2026

When OpenAI launched ChatGPT search on 31 October 2024, its announcement said the feature uses third-party search providers as well as content provided directly by its partners. It did not name the providers, and it still has not. Three things are nonetheless well established:

1. Independent tests in the weeks after launch found ChatGPT’s results closely tracking Bing’s index, including pages Google ranked differently and pages Bing had not indexed going missing from ChatGPT.
2. OpenAI has since been reported to be building and testing a search index of its own, which would reduce the dependence over time. It has not confirmed the details.
3. OAI-SearchBot crawls your site directly. Whatever the provider mix, OpenAI’s own crawler has to be able to read the page to summarise it.

The practical rule: treat Bing as the index you need to be in and OAI-SearchBot as the reader you need to allow. Both are cheap. If OpenAI’s own index takes over, nothing in the steps below becomes wrong; it just matters less which search engine you pinged.

Step 1. Allow the right bots, and verify the hits

User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /

If you also want to opt out of training, add a separate Disallow group for GPTBot; the three are evaluated independently. Then check the layer in front of robots.txt: a CDN or firewall rule that challenges non-browser requests will block OAI-SearchBot without ever reading your robots.txt. Add the ranges from openai.com/searchbot.json and openai.com/chatgpt-user.json to its allow list. The crawler list has the full set of user agents and the verification steps.

Step 2. Get your pages into Bing

Most sites have never looked at Bing. Three things close the gap in an afternoon:

• Bing Webmaster Tools. Free. Verify the site, submit the same sitemap you give Google, and check that your important pages show as indexed. Allow the bingbot user agent in robots.txt and in the firewall, the same way as in step 1.
• IndexNow. A push protocol Bing accepts: host a key file at the site root, then send each new or updated URL to the endpoint. One POST can carry up to 10,000 URLs, and the specification says a submission is shared with all participating engines, which include Bing, Yandex, Naver, Seznam, and Yep. Many CMS plugins do this automatically.
• Dates and titles. Bing, like any engine, needs a visible publication date and a title that names the thing. The same habits that help step 3 help here.

POST https://api.indexnow.org/indexnow
Content-Type: application/json; charset=utf-8

{
  "host": "example.com",
  "key": "<your-key>",
  "urlList": [
    "https://example.com/pricing",
    "https://example.com/docs/getting-started"
  ]
}

Step 3. Make each important page quotable

Both OAI-SearchBot and ChatGPT-User read the HTML as served. The model then lifts sentences from it. Three habits make a page the one it lifts from:

• One page, one question, answered first. The question in the heading, a two-sentence answer with the number and the date in the first paragraph, explanation afterwards.
• Text in the HTML. Prices in images, content behind tabs, and sections loaded by script after the page arrives are invisible to a plain fetch.
• Something citable. A figure with its source, a quotation, a dated reference. The GEO article covers the one controlled study on this and what it found.

Step 4. Keep facts consistent, and be realistic about llms.txt

A search-backed answer finds every copy of a fact you have published, including the 2024 blog post with the old price. Date the current page, annotate or redirect old copies, and make structured data match the visible text. The pricing article walks through the sweep.

On llms.txt: OpenAI’s crawler documentation does not mention it, and public log studies have counted almost no fetches from OpenAI’s agents. Publish it for the tools that read it, and do not expect it to move ChatGPT. The ChatGPT and llms.txt article has the evidence.

Step 5. Measure

1. Referrals. Filter analytics for utm_source=chatgpt.com. This is the cleanest signal any assistant gives you, because OpenAI adds it on purpose.
2. Logs. Count OAI-SearchBot and ChatGPT-User requests per week, verified against the published IP JSON. The first means you are being indexed; the second means a question caused a fetch.
3. A monthly audit. Ask ChatGPT your ten most important questions with search enabled, in a logged-out session, and record whether you appear in the sources and which page was cited. The citations article has the full routine.

When you are not cited

The first row is the one that fails silently. A checker fetches your robots.txt and pages the way a bot does and reports what OAI-SearchBot and the other common agents are allowed to read.

FAQ