ChatGPT gets its information from two main places. The first is what it learned in training, which stops at a cut-off date. The second is the live web, which it searches when a question needs fresh facts, using a search index and its own crawler. It can also use files you give it and, if you've turned it on, memory from earlier chats.
People ask me this most often after ChatGPT has said something about their business that's out of date or just wrong. The short answer is that it depends on which mode it's in. Once you know that, a lot of its odd behaviour makes sense.
Source one: what it learned in training
Before ChatGPT answered a single question, the model behind it was trained on a very large amount of text. That text came from a mix of publicly available web pages, books, licensed material and other data. From it, the model learned patterns in language and a broad base of general knowledge.
Training has a cut-off date. After that point, the model doesn't know about new events unless it looks them up. The cut-off differs between models, and OpenAI updates its models regularly, so I won't give you a date here. It would be out of date quickly.
Two things follow from this. First, the model isn't storing web pages like a library. It has learned from them, which is why it can sound confident and still get a detail wrong. Second, a small business that wasn't widely written about before the cut-off may get a thin, fuzzy answer, or none at all.
Source two: the live web, when ChatGPT searches
When a question needs fresh or specific information, ChatGPT can search the web and then write an answer from what it finds. This is the mode that matters most if you want your business to appear for questions like "who's the best accountant for freelancers?"
It doesn't browse the way you do. It sends a query to a search index, reads the most relevant pages that come back, and builds an answer from them, usually with links. OpenAI also runs its own crawler, called OAI-SearchBot, which helps ChatGPT's search find pages.
Whether it searches at all is down to the model, or to you if you switch search on. A question like "explain what a mortgage is" usually doesn't need it. A question about this week's news, or about a specific company, usually does.
Does ChatGPT use Google?
Not in the simple way people assume. ChatGPT search draws on third-party search providers, which include Microsoft Bing, along with OpenAI's own crawler, OAI-SearchBot. The exact mix isn't published and it can change, so be wary of anyone who tells you precisely how it works.
What I can say is this. Ranking well on Google doesn't automatically get you into ChatGPT's answers, and it doesn't rule it out either. In my experience, pages that are clear, easy to crawl and backed up by other sources on the web do well across all of these systems. I treat Google, Bing and ChatGPT as overlapping routes to the same goal, not separate games.
One practical check. Make sure your robots.txt file isn't blocking OAI-SearchBot by accident. OpenAI documents a separate crawler for training data, so blocking one doesn't have to mean blocking the other. If you've closed the search door without meaning to, you're invisible for no reason.
Files you give it, and memory
Two more inputs sit outside the wider web, and it helps to know they exist.
The first is files and pasted text. If you upload a document or paste text into the chat, ChatGPT can use it to answer. That's information you've supplied for that task, nothing more.
The second is memory. If you've turned it on, ChatGPT can carry details from earlier chats, like your job or your preferences, into later ones. That's about you, not about the world. It doesn't change what ChatGPT says about your business to anyone else.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
How you can tell which mode answered
You can usually tell from the answer itself. If it shows source links or small citation markers, ChatGPT searched the web, and you can click through to see which pages it used. If there are no sources, it's most likely answering from training, plus anything you've shared in the chat.
You can also ask it directly: "Did you search the web for that, and which sources did you use?" That isn't a perfect check, so trust the visible citations first. When an answer matters to me, I run the question twice, once with search on and once without, and compare. The gap tells you a lot about what the model already knows and what it needs to look up.
What this means if you want your business mentioned
Your business can appear through either route, but you influence them differently. Training is mostly out of your hands. The live web is where you can act today.
- Say plainly on your site what you do, who it's for and where you work, in the words a customer would use.
- Keep your key pages easy to crawl, and check your robots.txt file isn't blocking search crawlers by accident.
- Get your business described on other reputable sites, such as industry sites, directories and press, so more than one source agrees on who you are.
- Answer real customer questions on your pages in short, direct paragraphs.
- Keep your facts consistent everywhere: your name, your services and your contact details.
- Check what ChatGPT says about you every so often, with search on and with it off.
To see how it chooses between businesses once it has searched, read how ChatGPT decides which businesses to recommend. To test your own business, here's how to check whether ChatGPT is recommending you. The wider work of earning those mentions is what I call getting your business cited by AI.
What I've seen on my own site
I don't only write about this. I've been working on my own site's visibility in search and in AI assistants through 2026, and the numbers moved. Google clicks went from 1,322 in January 2026 to 6,412 in September 2026.
Visits from AI assistants went from 32 a month in January to 3,599 in September. Of those, 29 came from ChatGPT in January and 3,267 in September. I've written up what I changed in the full case study of my own site, so you can see the work behind the figures.
Curious what ChatGPT says about your business right now? I'm happy to look at it with you and talk through what I'd change. Here's how I can help.
If you want the full set, start at the ChatGPT Prompts and Limits: 30 Guides to Fun Prompts, Photo Edits and Daily Caps.
Frequently asked questions
Does ChatGPT use Google?
ChatGPT search draws on third-party search providers, including Microsoft Bing, and on OpenAI's own crawler, OAI-SearchBot. The exact mix isn't published, so I wouldn't assume Google is one of the sources.
Does ChatGPT have access to the internet?
Yes, but only when it searches the web. Without search, it answers from what it learned in training, plus anything you've shared in the chat.
Why does ChatGPT get facts about my business wrong?
Usually because little was written about your business before its training cut-off, or because sources disagree about your details. Clear, consistent information across your site and other reputable sites gives it better material when it searches.
Can I stop ChatGPT using my website?
You can control crawler access through your robots.txt file, and OpenAI documents separate crawlers for search and for training. Be careful, though. Blocking its search crawler can make it less likely that your pages are used and linked in answers.
