What is a context window?
Also: Token limit, context length
The context window is the maximum amount of text, measured in tokens, that a language model can read and use at once, including the question and any retrieved pages. Because it is limited, retrieval selects only the most relevant passages. Short, self-contained passages fit better than long pages with the answer buried.
Writing for it
Put the answer first, keep paragraphs focused and avoid filler before the point. Passages that need the rest of the page to make sense are harder to use.
Example: a model with a limited context window receives only the five most relevant passages of a long page, not the whole page.