ChatGPT · Platform deep dive
ChatGPT Search vs. Browsing Mode
ChatGPT now decides on its own, per message, whether to fetch live web results — there is no separate "browsing mode" toggle for most users anymore.
What actually happens
When you send a message, ChatGPT automatically detects whether it needs current, changing, or very specific information — like a recent event, a price, or a fact about a product — and if so, it runs one or more web queries in the background before answering. This is what OpenAI calls ChatGPT search: it is built into the default model experience for logged-in and logged-out users on Free, Plus, Team, Edu and Enterprise plans, not a separate mode you have to switch on. When search runs, the answer includes a sources panel or inline links you can click to verify where the information came from.
Base knowledge vs. retrieved knowledge
If ChatGPT decides a query is well served by what it already learned during training — stable facts, general concepts, coding help, creative writing — it answers directly without searching, and no sources appear. That means the same brand can show up two different ways in ChatGPT: as part of the model's general "background knowledge" from training data, and separately as a live citation pulled from a specific page at answer time. Both matter, but they are shaped by different things.
Why it matters for brand visibility
For the live-search path, being crawlable and indexed is what gets you considered at all — ChatGPT's search crawler is OAI-SearchBot, which is distinct from GPTBot (the crawler used for model training data), so your robots.txt needs to explicitly allow it if you want a chance at citation. For the background-knowledge path, there is no real-time lever: it reflects what was publicly documented about your brand at the time the model was trained, which is a much longer-horizon effort (press coverage, reviews, documentation, Wikipedia-style references).
Frequently asked questions
Does ChatGPT always search the web before answering?
No. It searches only when it judges the query needs current or specific information; for stable, well-established facts it answers from training data without showing sources.
How do I know if an answer used live search?
Look for a sources panel or inline citation links under or within the response. If there are no links, the answer likely came from training data alone.
Do I need to block or allow anything for ChatGPT to cite my site?
Your robots.txt should allow OAI-SearchBot specifically. Blocking it (even while allowing GPTBot) will keep your pages out of ChatGPT search results even if they were part of training data.