core
Context window
How many tokens of retrieved text an engine can feed its LLM at once — caps the number of pages it can cite per query.
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.