ChatGPT SEO: how to get your site found in ChatGPT
More people now put their questions straight into ChatGPT. When the answer triggers a web search, ChatGPT names sources and links them. "ChatGPT SEO" is the work of being one of them. Most of it is ordinary SEO. A few points are specific, starting with access for OpenAI's crawlers.
Key takeaways
- ChatGPT answers either from the model or by searching the web; only the search produces links.
- For search, OpenAI uses OAI-SearchBot; GPTBot is for training.
- The two are controlled separately in robots.txt.
- Nobody outside OpenAI knows the selection criteria. Work on what is documented.
How ChatGPT builds an answer
- From the model: knowledge from training data, with no source and a cut-off date.
- From a web search: ChatGPT fetches current pages, summarises them and links them.
Your site becomes visible and measurable mainly in the second case. Whether a search happens depends on the question: news, local matters, prices and product comparisons trigger one more often than general knowledge.
OpenAI's crawlers
OpenAI documents its crawlers and user agents publicly:
| User agent | Purpose | Controlled via robots.txt |
|---|---|---|
| OAI-SearchBot | Surfacing sites in ChatGPT's search features | Yes |
| GPTBot | Collecting content that may train OpenAI's models | Yes |
| ChatGPT-User | Fetching a page when a user triggers it in ChatGPT | OpenAI states robots.txt rules may not apply, since the user initiates it |
| OAI-AdsBot | Checking landing pages submitted as ads | Yes |
Swipe the table sideways to see all columns.
According to the documentation the settings are independent, so you can allow search and refuse training. OpenAI also notes that robots.txt changes take around 24 hours to take effect in its systems on the search side, and that ChatGPT-User does not determine search eligibility.
robots.txt: three typical setups
- Allow search, refuse training: `User-agent: OAI-SearchBot` with `Allow: /`, and `User-agent: GPTBot` with `Disallow: /`.
- Allow everything: no rules needed, as long as no blanket block exists.
- Refuse everything: `Disallow: /` for both. Your content then will not appear as a source in ChatGPT search.
Also check:
- CDN and firewall: some providers block AI crawlers by default, regardless of robots.txt.
- Server logs: look for the user agents and cross-check the IP addresses against the ranges OpenAI publishes, since user agents are easy to fake.
- Wildcard rules: a `User-agent: *` with a `Disallow` catches these crawlers too.
Content worth citing
OpenAI publishes no ranking factors. What follows from how it works:
- Text in the HTML: content loaded only by JavaScript is less reliable for a crawler.
- Answer first: address the question in the opening sentences of a section, details after.
- Clear structure: one question per section, descriptive subheadings, tables for comparisons.
- Concrete detail: figures with a source and a date, prices, steps, the area you cover.
- Your own experience: examples and data that are not everywhere else.
- Freshness: a visible update date and maintained content.
- Identifiable author and company: who writes, with what qualification, plus legal details and contact.
Presence beyond your own site
Answers usually draw on several sources. It helps if your business is described consistently elsewhere: directories and profiles with identical details, reviews on the platforms your sector uses, trade press mentions, and a decent position in conventional search, since AI search still leans on search indexes.
Measuring what arrives
- Referrer in analytics: visits from chatgpt.com can be isolated as a source.
- UTM parameters: ChatGPT appends tagging to many links; filter on it.
- Server logs: how often OAI-SearchBot fetches which pages.
- Regular sampling: ask your ten key customer questions each month and note who gets cited. Answers vary by user and moment, so read trends, not single cases.
And Google?
For Google's AI features the official line is that there are no extra requirements and SEO fundamentals still decide. Work done properly for ChatGPT generally improves your chances there too. Special files such as llms.txt are not needed for it.
Common mistakes
- Blocking AI crawlers, then wondering why you are invisible.
- Confusing GPTBot with OAI-SearchBot.
- Stuffing text with keywords instead of answering questions.
- Betting on supposed tricks nobody can evidence.
- Drawing conclusions from a single test.
Frequently asked questions
Is ChatGPT SEO different from normal SEO?
Mostly not. What is added is access for OpenAI's crawlers and how easily your content can be cited as an answer.
Do I need to allow GPTBot to appear in ChatGPT?
No. OpenAI states the search features use OAI-SearchBot; GPTBot concerns training.
How fast does a robots.txt change take effect?
OpenAI mentions roughly 24 hours for its systems to adjust on the search side.
Can I see how many visitors come from ChatGPT?
Yes, through referrer and UTM tagging in your analytics tool.
Does good optimisation guarantee a citation?
No. The criteria are unpublished and answers vary from one query to the next.
Sources
About this article. Produced by the MetaEX team with AI assistance, from the official documentation listed under Sources. Examples marked as illustrative are not client data. Last updated on 11 November 2026.
Want a read on your digital operation?
Talk to us: we first understand your situation, then tell you what makes sense.
Book a meeting