Training Corpus

What Training Corpus means for your business

A pre-training corpus is the material used in pre-training, the stage where the AI learns from a huge amount of text how words fit together. OpenAI lists preparing the data, pre-training and post-training as separate stages, and says different information may be used at each. It also says it increasingly uses synthetic data, material made by its own models, in some training.

For web pages, OpenAI says it uses only content that's freely open, doesn't knowingly collect from sources behind paywalls or on the dark web, and filters out spam, hate speech, adult content and sites that gather personal information. A site owner can block GPTBot in robots.txt, which OpenAI says tells it the content shouldn't be used for training. Our GPTBot entry covers the details. OpenAI also says its models don't keep copies of what they were trained on. Training adjusts the numbers inside the AI instead.

As we understand it, changing a page doesn't change what an AI already learned in training. When ChatGPT searches the web it can read the new page, but its trained knowledge stays the same until a newer version is trained. Since the corpus is mostly the open web plus partner data, we'd look at what's publicly readable about you, on your site and elsewhere, so whatever goes into the next one says the same thing. If you're not sure what the open web says about your business, checking what's written about you where AI companies can read it is where we'd begin.

Last reviewed October 29, 2026. All terms