How AI answers are made
Context window
The context window is the amount of text a language model can look at and use while it writes an answer, a kind of working memory.
In plain words
Anthropic describes it as the text a model can look back on and refer to while it writes. That is different from the huge amount of data the model was trained on.
Context windows are measured in tokens, the small pieces of text a model reads.
Example
On its models overview, Anthropic lists a context window of 1 million tokens for its current Claude models. That is room to read many long documents in one go.
Why it matters for AI search
Retrieved pages have to fit in the context window before the model can use them.
Even then, systems usually pass a selection of passages, not whole websites. In Anthropic’s own retrieval example, the model gets the top 20 passages.
What you can do
- Put the key answer early in each section, so it still works when only part of a page is used.
- Keep important facts in text, not only in images.
- Don’t pad pages: extra words don’t make a passage more useful.
Often confused with
- Training data
- Training data is what the model learned before release. The context window holds what it is reading right now.
- Knowledge cutoff
- A bigger context window doesn’t make the model’s built-in knowledge any newer.
Sources
Facts and sources last checked on 19 October 2026.
Related terms
How does your page do in AI search?
GEO-Score checks one page on 22 signals, from AI crawler access to structured data, and shows what to fix first.
Analyze a page