What is training data?

By the Variask team · Updated

Definition

Training data is the text a language model learned from before release, such as web pages, books and other sources. What the web said about your brand at that time shapes how the model describes you when it answers without searching.

How it works

  • Models learn patterns from very large collections of text.
  • Training ends at a knowledge cutoff date.
  • Answers without live search rely only on what was learned.
  • Newer models are trained on newer data.

What it means for brands

  • Consistent facts over time help future models describe you correctly.
  • For recent changes, rely on live search answers, which update faster.

Questions

Training data:
common questions

You can block some training crawlers, such as GPTBot or Google-Extended, for future training. That does not remove past data.

Not if you still allow the search crawlers, which are separate.

Related terms

Keep exploring

How AI search works

Knowledge cutoff

A knowledge cutoff is the date after which an AI model has no information from its training data. Anything that happened later is unknown to the model unless it searches the web or is given the facts.

Read

Technical

GPTBot

GPTBot is OpenAI's web crawler that collects publicly available pages that may be used to train OpenAI's models. It is separate from OAI-SearchBot, which powers search in ChatGPT.

Read

How AI search works

Large language model (LLM)

A large language model (LLM) is an AI model trained on very large amounts of text to understand and write language. LLMs power assistants such as ChatGPT, Gemini, Claude, Grok and DeepSeek.

Read
Full glossary