AI Visibility Academy · Chapter 08
ChatGPT: how it searches and cites web pages
ChatGPT answers some questions from what its model already knows and others after searching the web. OpenAI's help center says ChatGPT may search automatically when a question would benefit from current information, and that responses which use search may include citations you can open. To be one of those pages, yours must be reachable by OpenAI's search crawler and by the network in front of your site. OpenAI says placement is not guaranteed.
What does ChatGPT actually search for?
Not always the words the person typed. OpenAI says that when it works with outside search providers, ChatGPT typically rewrites the question into one or more targeted queries, and may send more specific ones after reading the first results. Saved memories and an approximate location from the person's IP address can shape the rewrite too.
Picture an e-bike repair shop in Portland. Someone asks for "a new battery for my commuter bike, who does that nearby?" The query that goes out may name a city and a battery type. A page that states the shop's city, the brands it services and a battery-swap price in plain text has something narrow to match. That is an inference, not a documented rule.
Which OpenAI bots decide if you can be cited?
As of October 2026, OpenAI documents four bots. Three bear on whether your pages can be cited; the fourth, OAI-AdsBot, only visits pages submitted as ads.
| Bot | Job | What robots.txt does |
|---|---|---|
| OAI-SearchBot | Surfaces sites in ChatGPT search | Opted-out sites aren't shown in search answers, but can still appear as navigational links |
| GPTBot | Crawls content that may train models | Disallowing says your content shouldn't be used for training |
| ChatGPT-User | Visits a page when a user's question needs it | OpenAI says robots.txt rules may not apply |
OpenAI says changes can take about 24 hours to apply. Its help center adds a second requirement: confirm that your host or CDN allows OpenAI's published searchbot IP addresses. A permissive robots.txt does nothing if a firewall refuses the request first; the lesson on CDN and firewall blocks shows where those blocks hide.
Product questions follow a separate path. OpenAI says shopping results are selected independently and are not ads, and Buying inside ChatGPT covers how products get in.
What isn't documented?
OpenAI doesn't say which factors rank one page above another, or by how much. It also warns that search results and citations can be incomplete, outdated or incorrect.
This week
Pick one buying question and ask ChatGPT three versions: the broad one, one with your city, and one with a constraint such as a budget. Open Sources on each answer and list every cited page, marking whether it is yours, a competitor's or a third party's. Then check that robots.txt allows OAI-SearchBot and your firewall passes its published IP addresses.