Training data
Training data is the text an AI model learned from before it was released. What it says about your brand is fixed until the next model.
Training data is everything an AI model learned from before it was released: web pages, books, articles, code, conversations. It's where the model's built-in knowledge comes from, including what it knows about your brand.
Once a model is trained, that knowledge stays the same until the company trains a new one. So when a platform answers from memory, it describes you as the web described you back then.
Two ways a platform knows about you
A platform can answer from what it learned in training, or it can search the web and read current pages first. Many do both. The training data gives the model its general picture of your category and the brands in it. The web search adds anything recent.
That's why a change you made last month can show up in one answer and not another. If the platform searched and found your new page, the answer is up to date. If it answered from memory, it isn't.
An example
For example, Fernhill Coffee started delivering across the UK eight months ago, after years of delivering only within London. A model trained before the change has read hundreds of pages saying "London only." When a platform using that model answers without searching, it repeats the old fact.
Fernhill can't change what the model was trained on. It can make sure the new fact is on clear, easy-to-find pages, on its own site and on the review sites and directories the platforms search, so answers that do search get it right. And next time the model is trained, those pages are part of what it reads.
Can you keep your pages out?
Most AI companies run a separate bot to collect pages for training, and you can block it in your robots.txt. That keeps your new pages out of future training. It doesn't remove what's already been learned, and it doesn't stop the search bots that read your pages for answers, unless you block those too.
In AnswerOrbit
When an answer says something wrong about you without using any page, Inaccuracy drivers in Claims lists the platform's own knowledge as the source. That's usually a sign the mistake comes from training. The Audit tab in Website and AI traffic shows which training bots and search bots your robots.txt lets in.