ChatGPT SEO: how to rank in ChatGPT, and how it recommends brands

Where does ChatGPT get its web results?

ChatGPT doesn’t go to the web for every question. According to OpenAI’s help page, it may search automatically when a question would benefit from current information, and users can start a search themselves by choosing Search from the tools menu. Answers that use web search can include citations, and selecting Sources shows the cited sources along with other relevant links.

The same page says ChatGPT search sometimes partners with other search providers. When it does, it typically rewrites the user’s question into one or more targeted queries and sends them to those providers; after reviewing the first results, it may send more specific queries. The privacy policies OpenAI points to for these providers are Microsoft’s and Shopify’s.

That is as far as we can verify the Bing question. We couldn’t find a statement in OpenAI’s current documentation along the lines of “ChatGPT uses Bing’s rankings one to one”. So we don’t treat ranking well in Bing as a guarantee of appearing in ChatGPT; adding your site to Bing Webmaster Tools, though, stays on our list as a low-effort housekeeping step.

OpenAI is also clear about ranking: ChatGPT ranks search results using multiple factors intended to help users find relevant, reliable information, and placement is not guaranteed.

OAI-SearchBot, GPTBot and ChatGPT-User: what does each one do?

OpenAI uses separate user agents, and in robots.txt each one is managed independently of the others. For ChatGPT SEO, the line that matters most is OAI-SearchBot. That independence means you can block GPTBot and allow OAI-SearchBot: you signal that you don’t want your content used for model training and still appear in ChatGPT search. OpenAI says it can take about 24 hours for a robots.txt change to reach its search systems.

robots.txt isn’t the only gate. OpenAI’s help page also says your website host or content delivery network has to allow traffic from OpenAI’s published search crawler IP addresses. A firewall with strict bot protection can stop the crawler at the door even when robots.txt lets it in. Check which response codes OAI-SearchBot requests get in your server logs.

According to OpenAI’s documentation, the three crawlers do the following (a fourth, OAI-AdsBot, only visits landing pages submitted as ads in ChatGPT):

• OAI-SearchBot: used to surface websites in ChatGPT’s search features. Content from sites that block it is not shown in ChatGPT search answers, though the site can still appear as a navigational link.
• GPTBot: crawls content that may be used to train OpenAI’s foundation models. Blocking it indicates that you don’t want your content used for training.
• ChatGPT-User: visits pages for actions a user starts in ChatGPT. It doesn’t crawl the web automatically, and it isn’t used to decide whether content can appear in search.

Why does ChatGPT recommend one brand over another?

OpenAI doesn’t publish a document explaining why a brand is recommended in an answer. What is known is that answers using web search are compiled from the sources found and cite those sources. So what those sources say largely decides how your brand is described.

Those sources aren’t only your own site. Comparison lists, industry guides, directories, news and customer reviews can all feed into the answer. Describing clearly and consistently on your own site what you do and who you do it for is the first step towards the same description settling into external sources.

A practical place to start: ask ChatGPT, using your brand name, “What does X do, and who does it serve?”, then open the sources the answer cites one by one. If your service is described in an outdated, incomplete or wrong way, fix it at the source, not in the answer: the description on your own site, your Google Business Profile, your LinkedIn page, your listings in industry directories. Ask the same question again a few weeks later and record the change.

It is worth separating two outcomes here: your page being cited and your brand being recommended. Your guide can be the source while the recommendation goes to a competitor; we explain how to track the two separately in our guide to GEO, SEO and AEO. Don’t spend budget on collecting artificial “mentions” either: Google writes in its guide to AI search that seeking inauthentic mentions isn’t as helpful as it might seem.

How should you structure content so ChatGPT can quote it?

None of the following is a ChatGPT-specific trick; they are changes that make the same page work better on Google and for readers too. It is also why the guides on this site open with a “Short answer” box: neither the reader nor a system writing an answer should have to hunt for the main point.

Service businesses shouldn’t skip pricing and scope either. If you don’t want to publish exact figures, explain what the price depends on and how the process works. If the answer to the question customers ask before buying, “what does the price depend on?”, isn’t on your page, the answer has to take that information from another source.

• Short answer first: answer the page’s main question in two or three sentences in the first paragraph, in a way that still makes sense when it is lifted out on its own.
• One clear service definition: state what you do, who for and in which market, in the same words on every page.
• Sources: when you cite a platform rule, a statistic or a regulation, link to the primary source.
• Author and date: make it visible who wrote the page, what their expertise is and when it was last updated.
• Information you can’t find elsewhere: add details such as scope, process, deliverables and worked examples.
• Readable HTML: having the main text in the page source is the safest route; OpenAI’s crawler documentation makes no commitment about rendering JavaScript.

Does llms.txt help you show up in ChatGPT?

llms.txt is a file proposed by Jeremy Howard in 2024 that summarises a site’s important pages for language models in Markdown. Even the project’s own site describes it not as a standard but as a proposal to standardise.

The conditions OpenAI gives for appearing in ChatGPT search come down to allowing OAI-SearchBot and keeping traffic from its published IP addresses open; llms.txt isn’t among them. Google, for its part, says plainly that you don’t need such files to appear in Search and that Google Search itself doesn’t use them.

Our view: preparing an llms.txt file is a few hours’ work, and for some tools it can act as a tidy index of your content. But it isn’t a lever for ChatGPT visibility; it guarantees neither rankings nor citations. Put the budget into the pages themselves first. We cover where llms.txt sits in GEO work in our GEO vs SEO vs AEO guide.

How do you measure traffic from ChatGPT?

Links from ChatGPT often carry the parameter utm_source=chatgpt.com, and GA4 shows that value as the session source. OpenAI’s FAQ for publishers also says ChatGPT automatically adds this parameter to referral URLs. Even so, don’t rely on the parameter alone; check the referring domain as well.

When you read server logs, don’t trust the user agent name alone, because it can be spoofed. OpenAI publishes the IP addresses its search crawler uses. Compare the OAI-SearchBot requests in your logs with that list to tell the real crawler from imitators, and use the same list when you let it through your firewall.

We set up measurement in three layers. Each layer isolates the problem of the one before it: without crawling there is no visibility, and without visibility there is no traffic.

• Crawling: check in your server logs which pages OAI-SearchBot visits and whether it gets errors. If an important page never gets a visit, the problem is access before it is content.
• Visibility: write the 10–20 questions your customers ask into a fixed list and ask them in ChatGPT at set intervals with exactly the same wording; record whether you are cited and whether your brand is mentioned.
• Traffic and enquiries: track visits with chatgpt.com as the session source separately in GA4 and check whether they turn into forms, calls or purchases.

Can you use AI for SEO? What does Google say?

The second meaning of “AI SEO” is producing content with AI tools. Google’s position has been the same since February 2023: its ranking systems aim to reward original, high-quality content, however it is produced. Appropriate use of AI or automation is not against its guidelines.

The line is drawn at purpose. Using automation, including AI, to generate content with the primary purpose of manipulating search rankings violates Google’s spam policies. One of the examples in Google’s definition of scaled content abuse is exactly this: using generative AI tools to generate many pages without adding value for users.

So the question is not “was AI used?” but “does this page give the reader something they can’t find elsewhere?”. Hundreds of pages produced from one template by swapping the city, industry or product name fail that test however they were written. A page that covers one topic in depth, with a worked example and sources, can pass it even if it was prepared with AI’s help.

Google also suggests asking Who, How and Why. Give accurate author information where readers would expect it, and consider disclosing the use of AI or automation where readers might wonder “how was this created?”. Using AI for research, drafts and checklists while keeping the figures, examples and final decision with a person is a division of labour that fits this line. We draw the same distinction on the advertising side in our guide to creating ads with AI.