ChatGPT SEO: how ChatGPT picks sources, and the work that gets you named

ChatGPT SEO is the work of getting your business named and cited when somebody asks ChatGPT a buying question. It rests on one fact about how ChatGPT answers: for many questions it runs a search and writes from pages it just fetched, and for others it writes from what the model already holds. The searched answers are the ones you can influence this month, and three things decide whether you appear in them. Whether your site lets OpenAI's named crawlers in, whether any single page of yours answers the question outright, and whether other sites say the same thing about you.

OpenAI's crawlers, and what each one controls

OpenAI publishes its crawler list and the robots.txt tokens for each one. This matters because a robots file written for Googlebot years ago says nothing about these agents, and a default deny rule at your CDN can be quietly keeping you out of ChatGPT's search results.

User agent	What it is for	What blocking it does
OAI-SearchBot	Surfaces websites in ChatGPT's search features	OpenAI states that opted out sites will not be shown in ChatGPT search answers, though they can still appear as navigational links
GPTBot	Crawls content that may be used to train OpenAI's foundation models	Signals that your content should not be used for training. It does not remove you from search
ChatGPT-User	Fetches a page because a person or a custom GPT asked for it	OpenAI notes this is not automated crawling, that robots rules may not apply, and that it is not used to decide what appears in search
OAI-AdsBot	Checks landing pages submitted as ads on ChatGPT	Only visits pages submitted as ads

The settings are independent, which is the useful part. You can be in ChatGPT's search results while refusing to be training data. OpenAI's current documentation is at platform.openai.com/docs/bots, and it says a robots.txt change takes roughly 24 hours to register on their side.

The robots block to copy

User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /

User-agent: GPTBot
Allow: /

User-agent: *
Disallow: /wp-admin/
Disallow: /cart
Sitemap: https://example.com/sitemap.xml

To stay in ChatGPT search while declining training, change the GPTBot rule to Disallow: / and leave the other two as they are. Then confirm the real behaviour in your server log rather than in the file, because a CDN rule or a bot management product can override what robots.txt says. Full walkthrough in which AI crawlers should you allow.

When ChatGPT searches, and when it does not

The decision is made per question. Questions about current prices, recent events, named local businesses, and direct comparisons tend to trigger a search. Broad category questions often come back from the model itself, with no sources attached.

You can settle it in one move: ask your buyer's question, then ask ChatGPT whether it searched the web and which pages it read. It will tell you, and the answer decides which half of the work you do next. We tested the pattern in does ChatGPT use live search or memory.

How a page gets cited

Four properties keep showing up in the pages ChatGPT quotes, and all four are things you control.

• The page answers one question. Retrieval lifts a passage, not a site. A page whose title is the question, with the answer in the first two sentences, hands over a clean passage. A services page covering nine things hands over nothing.
• The facts are specific and dated. A count, a price with the month you checked it, a named source. Specifics are what a model will repeat, because a vague phrase gives it nothing to stand on.
• The text is in the HTML. If your key sentence only appears after JavaScript runs, treat it as a sentence the fetcher may never see. View source and search for it.
• Somebody else says the same thing. A claim that exists only on your own domain has one source behind it. The same claim on a directory, a review platform and a trade publication has four.

We went deeper on the selection behaviour in how ChatGPT picks which companies to name and on the source mix in where ChatGPT gets its sources.

How to test it properly

ChatGPT has features that will flatter you if you let them, so test under controlled conditions.

• Use a temporary chat, or turn memory off. Memory, custom instructions and the history of your own chats all shape what comes back. Your own account is the least representative place to run a test.
• Keep your company name out of the question. Asking "is our firm good for this" primes the answer. Ask the question your buyer would ask and see who appears. The fair protocol is in should you ask ChatGPT about your own company.
• Run each question more than once. The same question can return different companies on two runs, so record a range across runs rather than one reading.
• Save the citations, not just the verdict. The cited sources tell you which sites to go and get listed on next. That list is the most actionable output of the whole exercise.
• Test more than one engine. ChatGPT is one of five surfaces your buyers use, and the answers differ. See why one AI app is not enough to measure with.

Everything above is about how the sources get chosen. The question that follows is whether you are one of them, and for which buyer questions, which no analytics tool reports. AI Knows Us asks a fixed set of buyer questions without naming the business, records whether you were named and what was cited instead, and then does the page and listing work that changes the answer.

This page is about ChatGPT specifically. For the whole method across engines, see LLMO and how it relates to GEO, AEO and SEO. For a step by step check on your own business, see how to check whether ChatGPT knows your business.

Questions buyers ask about ChatGPT SEO

Does traditional SEO help me in ChatGPT?

The technical half carries over almost completely, because a page that cannot be crawled or rendered is out of both games. The content half changes: ranking tenth still puts you on a results page, while being the tenth company ChatGPT might have mentioned means it mentioned nobody but your competitors.

Can I pay to appear in ChatGPT's answers?

OpenAI runs advertising on ChatGPT, and it documents OAI-AdsBot as the agent that checks submitted landing pages. That is a separate, labelled placement from the organic answer. Nobody sells a place inside the organic answer, so treat any such offer as a reason to ask for the contract.

How long until a new page shows up?

Once the page is crawlable and indexed, searched answers can start using it within weeks. Answers written from model training move on the timescale of the next model, which is why third party presence is the long game.

Why does ChatGPT name my competitor and not me?

Usually one of three reasons: its fetcher cannot read your site, you have no page that answers that specific question, or nothing outside your own domain corroborates you. Check them in that order, because the first is the cheapest to fix.

What if ChatGPT says something wrong about my business?

Correct the source rather than arguing with the output. Find which pages carry the wrong fact, fix them at origin, and publish a page that states the correct version plainly with a date. The sequence is in what to do when ChatGPT gets a fact about you wrong.

What to do first

• Open your robots.txt and look for OAI-SearchBot, GPTBot and ChatGPT-User by name. Add the block above if they are missing.
• Grep your access log for those agents. If none of them have visited, the problem is access, not content.
• In a temporary chat, ask the five questions your buyers ask, without naming your company. Save the answers and the cited sources.
• Write one page for each question where a competitor was named and you were not.
• Pick the three sites ChatGPT cited most often and get yourself properly listed on them.