Where does ChatGPT get its information about businesses?

Where ChatGPT gets its information: two layers

When someone asks ChatGPT about a business, the answer can come from two different layers. They behave differently, and mixing them up is the source of most confusion.

1. Training. The model learned patterns from a large body of text before you ever typed a question.
2. Live search. For questions that need current information, ChatGPT can search the web and base the answer on pages and data it finds.

For a local business owner, the second layer is the one you can influence this month. It is also the one OpenAI describes least. This article separates what OpenAI confirms from what is only reported, as part of our wider look at AI search optimization.

Layer 1: training data

OpenAI says its foundation models are developed using three primary sources: information that is publicly available on the internet, information it partners with third parties to access, and information that users, human trainers and researchers provide or generate.

For public web content, OpenAI says it uses only information that is freely and openly accessible. That can include public webpages, forums and blogs. It says it filters out material such as spam and sites that aggregate personal information. Website owners can use robots.txt to disallow GPTBot, the crawler OpenAI describes as one that may crawl public content to help train its models.

Two things follow for a business.

• A model does not look you up. OpenAI says the models do not store copies of what they train on. They adjust internal values to reflect patterns, then generate new text. So a model answering from training alone is working from patterns it learned about your kind of business, and possibly about you if you appear often on the public web.
• Training is not current. Training happens in stages before a model is released, so a price you changed last week cannot be in it. That is our reading of how the process works, and it is the reason live search matters.

Blocking GPTBot is a separate decision from appearing in search. OpenAI documents the two as independent settings, which we cover below.

Layer 2: live web search

OpenAI’s help page says ChatGPT may search the web automatically when a question would benefit from current information. People can also start a search by hand. The Enterprise and Edu help page describes the steps: ChatGPT can turn the request into one or more search queries, retrieve relevant results, and generate an answer with links to sources.

Here is what OpenAI confirms about the sources behind that search:

• It sometimes uses other search providers. OpenAI says ChatGPT search sometimes partners with other search providers, and typically rewrites the question into one or more targeted queries that it sends to them. It may send follow-up queries after reviewing the first results.
• Two providers are named in privacy links. For general use, the help page points readers to Microsoft’s privacy statement and Shopify’s privacy policy for how those providers process queries. It does not list which providers answer which questions.
• Bing is named for Enterprise and Edu. The Enterprise and Edu page says ChatGPT may share disassociated search queries with the Bing search engine to return web results. It also says structured data from a prompt may be shared with Bing or third-party data providers for specialized results such as weather, sports and finance.
• It has its own search crawler. OAI-SearchBot “is used to surface websites in search results in ChatGPT’s search features.” OpenAI says to allow it in robots.txt to appear in ChatGPT search answers.
• It uses location. OpenAI says ChatGPT may use an approximate location from the IP address, and may share location with trusted third-party providers that provide local information such as nearby business listings.
• It shows reservation partners for restaurants. For restaurants, the help page says availability may come from OpenTable, Resy or Yelp. That is one example of how specialized data can enter an answer.
• It cites sources. Responses that use web search may include citations, and a Sources list when available. OpenAI’s developer docs add that the sources list can be longer than the citations shown.
• Placement is not guaranteed. OpenAI says ChatGPT ranks results using multiple factors and that placement is not guaranteed.

Does ChatGPT use Bing? Does it use Google?

These are the two most common questions, and the honest answer is a table of what is confirmed and what is not.

The practical conclusion: you cannot target “the ChatGPT index.” You can make your information available and consistent in the places any of these systems read. That list is short: your own website, the search engines (Google and Bing), and the business directories and profiles that supply local listings.

Why consistent listings matter

If a search system reads several sources and they disagree, something has to be chosen or skipped. Disagreements are easy to create: an old phone number on a directory, “St.” vs “Street,” a closed location still listed, hours that changed in one place only.

Here is the case for consistency, using only what the sources say.

• OpenAI says ChatGPT may share location with third-party providers that supply nearby business listings. Those listings come from providers, not from your website alone.
• Google says complete, accurate business information makes a Business Profile more likely to appear in local results, and that inaccurate information can keep it from showing for relevant searches.
• Microsoft’s Bing guidance for AI visibility says to keep address, hours and contact details current through Bing Places for Business, and to align text, images and video so they describe the same entities. It also recommends updating content regularly so AI systems reference the current version.

Add them up and the advice is clear. Do not guess which listing matters most. Make them all agree: business name, address, phone, hours, website address and the list of services. We cover the full sequence, with a checklist, in how to show up in ChatGPT.

What this means for your business

You cannot see which source ChatGPT used for a given answer unless it shows citations. But you can act on what is within reach:

1. Let OAI-SearchBot crawl your site. Check robots.txt and your host or CDN settings. OpenAI says a robots.txt change can take about 24 hours to take effect.
2. Decide about GPTBot on its own merits. Disallowing GPTBot signals that your content should not be used to train models. OpenAI says it does not affect OAI-SearchBot, and each setting works independently.
3. Put the facts in text on your pages. Services, area, prices, hours, how to book. If the page does not say it, a search provider cannot quote it.
4. Make your profiles match. Google Business Profile, Bing Places, Apple Business Connect, Yelp and your social pages.
5. Date the pages. A visible “Updated” date helps customers and shows which version is current.
6. Test on a schedule. Ask the same questions every month and write down what each assistant says and cites.

For Google’s side of this, see Google AI Overviews. For how SEO, GEO and AEO relate, see GEO vs SEO vs AEO.

How to see where an answer came from

Open the citations. OpenAI says responses that use web search may include citations you can select, and a Sources view shows cited sources and other relevant links. Check each source for old or wrong facts about your business.

If a result looks wrong, OpenAI’s help page suggests asking ChatGPT to search again using a specific source, date or location. If local results show the wrong area, include your city, neighborhood or postal code in the question. A VPN can change the approximate location, so test the way a customer would.

Expect variation. OpenAI says the same question may yield different answers across different queries, because of an inherent element of randomness in how the model responds. If your business is missing once, it proves little. If it is missing or wrong in most runs, look at the sources it did cite and fix what they say.

What nobody controls

We cannot control what ChatGPT, Google AI Overviews, AI Mode or Perplexity recommend, and nobody can. OpenAI does not publish a list of all its sources or how it weighs them, and it says placement is not guaranteed. What you control is the quality, consistency and reachability of your own information. That is also what we check in our GEO/AEO audit, which is from $750 and takes 1-2 weeks. It ends with a record of what AI answers say about you and a re-check after the fixes. If you would like to see where you stand first, ask for a free demo.

Published October 25, 2026. Updated October 25, 2026.