ChatGPT SEO: how we get your brand named and cited in ChatGPT

ChatGPT can answer from what the model learned in training, or it can run a web search and cite what it finds. When it searches, it shows links to the sources. That second path is the one you can influence today, because it depends on pages that exist on the open web right now.

OpenAI documents four user agents on its developer site (read on 15 October 2026). Three matter for visibility and training choices; the fourth checks landing pages submitted as ChatGPT ads and is outside the scope of organic work. OAI-SearchBot is used to surface websites in ChatGPT search features. GPTBot is used to collect content that may train OpenAI's foundation models. ChatGPT-User is used when a person asks ChatGPT something that needs a live page visit, for example inside a conversation or a custom GPT. As of October 2026, OpenAI says each of these settings is independent, so a site can allow search and still disallow training.

OpenAI also states that a site that opts out of OAI-SearchBot will not be shown in ChatGPT search answers, though it can still appear as a navigational link. The documentation recommends allowing OAI-SearchBot in robots.txt and allowing OpenAI's published IP ranges. It says robots.txt changes can take about 24 hours to be picked up for search.

• OAI-SearchBot is not blocked in robots.txt
• Your firewall or CDN does not challenge OpenAI's published IP ranges
• GPTBot is a separate choice: allow or block it for training reasons, not for visibility reasons
• ChatGPT-User may not follow robots.txt, because a person asked for the visit

ChatGPT answers change between runs. The same prompt asked twice can cite different sources. Answers also differ by location, by logged in state, by memory settings and by model version. That is why a single screenshot proves very little, and why we never treat one good answer as a result.

What is observable is a pattern. Across many prompts and repeated runs, certain things show up again: pages that answer the question in the first lines, brands that are described the same way on their own site and on third party sites, and domains that are easy for crawlers to reach. We treat those as strong hypotheses, not laws. Engines change often, and what works in October 2026 may be weaker in 2027.

This is the honest frame for every ChatGPT project we run. We fix what is documented and what is plainly broken. We improve what is likely to help. We test, and we tell you what moved and what did not.

We keep the prompt list fixed. We run it before we change anything and again after the fixes, at the same settings where we can. We record whether your brand is named, whether your domain is cited, which page is cited, and whether the description is correct. We note every run date, because a result without a date means little.

We also look at traffic. ChatGPT search links often arrive in analytics with a utm_source of chatgpt.com, which we have seen in client data but have not verified in OpenAI documentation, so your analytics may show visits from that source. Many answers never produce a click, so a mention can matter even when the visit count stays low. Our guide to measuring GEO performance explains how to read both signals.

We do not report a ranking, because ChatGPT does not have a stable one. We report presence and accuracy across a defined set of prompts.

A common safe pattern is to name OAI-SearchBot with an Allow rule for the public site, then decide GPTBot separately. In plain terms the file has a group that starts with User-agent: OAI-SearchBot followed by Allow: /, and a second group that starts with User-agent: GPTBot followed by either Allow: / or Disallow: /, depending on your training choice. Put private paths such as admin, cart and account pages under Disallow in the group that applies to all agents.

Order and specificity matter. A crawler follows the most specific user agent group that matches it. If you name OAI-SearchBot explicitly, it will follow that group and ignore the wildcard group. That is why a named Allow works even when the wildcard group is strict. Our guide to AI crawlers and robots.txt goes through more patterns, and the robots.txt generator builds a file for you to review.

After you publish a change, OpenAI says it can take about 24 hours for search to reflect it. So plan your first retest for the next day, not the same hour. If you use a CDN with cached robots.txt, purge the cache first, or the old file may keep being served.

• Named group for OAI-SearchBot with Allow for public pages
• GPTBot group set deliberately, with the reason written down
• Private paths blocked for all agents
• Robots.txt cache purged after each edit
• Retest scheduled at least one day later

Tracking your brand name alone tells you very little. Buyers rarely ask for you by name before they know you. They ask for a category, a comparison or a problem. A useful ChatGPT prompt set mixes four types: category prompts (best [service] in [city]), comparison prompts ([option A] versus [option B] for [use case]), problem prompts (how do I fix [issue] in [situation]) and brand prompts (is [brand] good, what does [brand] charge). Each type exercises a different part of retrieval.

Write prompts the way a buyer types them, not the way a marketer would. Include the city or country when the purchase is local, and the buyer's role when it is B2B. In the Pilot we use up to 20 prompts and keep them fixed from the baseline to the retest, so that any change is comparable. If you want to build a list yourself, start with our guide to researching buyer prompts.

Run each prompt more than once. ChatGPT answers vary between runs, so a single answer is an anecdote. Record the date, whether search was triggered, the brands named, the domains linked and any wrong statement. Search is not triggered for every prompt, and a prompt that does not search tells you about model memory instead of your site.

Example, hypothetical: a Dubai dental clinic with a good site finds that ChatGPT never names it for "best pediatric dentist in Dubai". The audit shows three problems. The firewall returns a challenge page to unknown bots, so OAI-SearchBot cannot fetch the pages. The pediatric service sits inside a long page about all treatments, with no heading that matches the question. And two old directory listings show a previous clinic address.

The fixes follow the problems. We ask for the bot protection rule to be relaxed for OpenAI's published ranges. We create a dedicated pediatric dentistry page that opens with a two sentence answer, lists treatments, states the languages spoken and the area served, and adds an FAQ built from questions parents actually ask. We align the address across the site, the Google Business Profile and the main directories. We add Organization and local business data that matches the visible text.

What we would not claim: that this guarantees a mention. After the changes we run the same prompts again and report what changed, including if nothing did. If the clinic is named in some runs and not others, we say so. For the sector view, see GEO for clinics and local GEO.

ChatGPT is one of five engines in the Pilot baseline, and most of the work helps all five. The crawler and entity fixes are shared. What is specific to ChatGPT is the OpenAI crawler rules, the prompt set and the retest. If you only want to know where you stand today, the free AI visibility checker is a starting point.

The GEO Pilot is a fixed $999 (USD), one time price, delivered inside 30 days, done for you. It covers one domain and one language, and it does not include a new site build, ongoing digital PR, link building or paid media. Read the full scope on the GEO Pilot page, or compare with the wider GEO overview.