Arrow 
When ChatGPT needs current information it searches: OpenAI’s crawler (OAI-SearchBot) and third-party search providers find pages it can cite. When a user shares a link, ChatGPT-User opens that page on request. Everything else comes from training, collected before a cutoff date by crawlers such as GPTBot and through licensed data. To appear in answers, make sure OAI-SearchBot can reach you, that your pages are indexed in Bing, and that your facts are consistent everywhere.
Three paths, three different controls
The most common mistake is to treat “ChatGPT” as one system. OpenAI documents separate agents for separate jobs, and each one follows its own rules. That is why a site can block training and still appear in ChatGPT search, or the reverse.
| Path | What it does | Agent | Can robots.txt control it? |
|---|---|---|---|
| Live search | Finds current pages to cite in ChatGPT search answers | OAI-SearchBot, plus third-party search providers and partner content | Yes, for OAI-SearchBot |
| User-requested fetch | Opens a page a user shares or asks it to visit | ChatGPT-User | OpenAI says robots.txt rules may not apply to user-initiated requests |
| Training | Content that may be used to train future models | GPTBot (among other data sources) | Yes, for GPTBot — but it does not control search visibility |
OpenAI’s publisher FAQ is explicit on the practical point: to be included in summaries and snippets in ChatGPT search, don’t block OAI-SearchBot. To keep content out of potential training, disallow GPTBot. The two settings are independent.
Does ChatGPT use Google or Bing?
ChatGPT search combines OpenAI’s own crawler with third-party search providers and content from partners. OpenAI’s help documentation on searching the web with ChatGPT describes this mix; Microsoft Bing is the provider most often named. OpenAI has not published a complete list of providers, so articles claiming ChatGPT “uses Google” rely on outside experiments rather than documentation.
The practical consequence is simple. A strong Google ranking helps because it usually reflects a useful, crawlable page, but it does not transfer automatically. Check two things separately:
- Bing indexing. Verify your site in Bing Webmaster Tools, inspect your key URLs and use IndexNow when pages change. Bing’s new AI Performance report shows citations in Microsoft Copilot and Bing’s AI summaries — a related signal, not ChatGPT data.
- OAI-SearchBot access. Your robots.txt, firewall and CDN must let it through. Our guide on which AI crawlers to allow has copyable configurations.
Where do training-based answers come from?
When ChatGPT answers without searching, it relies on what the model learned from data collected before its training cutoff. GPTBot is OpenAI’s training crawler, and OpenAI also licenses data — for example its partnership with Reddit announced in May 2024, which gives it access to Reddit’s data API.
Training-based knowledge ages. If a model learned your old prices, your former brand name or a service you stopped offering, it can repeat them until a search result or a newer model corrects it. We cover those situations in why AI shows your old prices and why ChatGPT gets your business wrong.
What should a business check first?
Start with access, then clarity, then evidence. In our 25 September 2026 check of 100 yachting websites in Miami and South Florida, only 2 of the 78 sites with a robots.txt file blocked any AI crawler. Access is rarely the whole problem; being the clearest, best-supported source usually is. Our Miami yachting index documents the sample.
- Access: OAI-SearchBot allowed in robots.txt, and no firewall rule returning 403 to legitimate bots.
- Indexing: your important pages indexed in Bing, with a current sitemap.
- Clarity: one consistent description of what you do, where and for whom, on your site and your profiles.
- Evidence: reviews, press, community discussions and directories that describe you accurately — see why ChatGPT cites Reddit.
- Measurement: ChatGPT adds
utm_source=chatgpt.comto the links it cites; GA4 now groups recognised assistants in an AI Assistant channel.
How long do changes take to show up?
Robots.txt changes are fast; knowledge changes are not. OpenAI says it can take about 24 hours after a robots.txt update for its systems to adjust. New or corrected pages appear in search answers once they are recrawled and judged relevant. What a model learned in training only changes with future models, which is why a clear, current page that search can find matters more than trying to “update ChatGPT”.
Sources and editorial scope
This guide explains documented behaviour of ChatGPT’s search, fetching and training agents as of September 2026. OpenAI does not publish its full list of search providers or how answers are ranked, so we separate documented facts from outside claims.
- OpenAI — Overview of OpenAI crawlers
- OpenAI Help Center — Publishers and Developers FAQ
- OpenAI Help Center — Searching the web with ChatGPT
- Bing Webmaster Blog — AI Performance in Bing Webmaster Tools (public preview, February 2026)
- Fortune — OpenAI partners with Reddit (May 16, 2024)
- Arrow AI — Miami & South Florida yachting website readiness index
The Arrow read
Being reachable is not the same as being chosen.
Most businesses we review are not blocked from ChatGPT. They are simply not the clearest source for the questions their buyers ask. Start with what AI says about your business today, then fix the facts and evidence behind it. Our free audit checks access, indexing and consistency in one pass.