Glossary entry context-windowCore concepts

Context window

How many tokens of retrieved text an engine can feed its LLM at once, which caps the number of pages it can cite per query.

Filed under: Core concepts
In the rubric: Information density

Definition

Context window is the maximum input size of an LLM measured in tokens (~3-4 chars each). Claude 3.5 = 200k, GPT-4 Turbo = 128k, Gemini 1.5 Pro = 2M. The context window caps how many retrieved pages fit per query: a 200k context engine can fit ~30 average pages, a 2M context engine ~300. Bigger contexts = more citations per query = your page faces more competition.

Example

Claude with 200k context can fit ~30 pages per query; if 5 are cited, your page competes against 25 others to be one of the 5.

How to optimize

Bigger contexts favor concise, dense pages: engines preferentially cite pages that pack max information per token. Cut fluff, lead with substance.

Check a page against a real AI answer

Claude’s live answer to your buyer’s question, its sources, your readiness score and the fixes. Free, no signup, about half a minute.

Run a free audit