context window
Also called: context length, context size, context window size
A context window is the maximum amount of text, measured in tokens, that a large language model can consider at once. It covers both the input (the prompt plus any retrieved documents) and the output the model generates, and it acts as the model's working memory for a single request.
Context windows are counted in tokens, the sub-word units a model’s tokenizer produces (roughly four characters, or about three-quarters of a word). Everything in a request shares that budget: the system prompt, the conversation history, any retrieved documents or images, tool definitions, and the response the model writes back. Sizes vary widely, from 8,000 tokens on older systems to 1 million tokens on current frontier models like Claude and Gemini.
Bigger is not automatically better
Anthropic documents a failure mode it calls “context rot”: as the token count climbs, recall and accuracy drop because a transformer has to weigh every token against every other one. Google reports single-fact retrieval near 99% accuracy, but pulling many separate details from a full window at once gets less reliable. The practical rule is to load the smallest set of high-signal tokens that does the job, not everything you have.
When input passes the limit, requests fail or truncate. On Claude, input past the ceiling returns a prompt is too long 400 error, and overflow during generation stops with model_context_window_exceeded. Retrieval-augmented generation (RAG) exists partly to work around small windows by fetching only the relevant passages instead of stuffing in the whole corpus. Even with million-token windows, curation still beats brute force.
How it affects your traffic
When an AI assistant answers a query, your page is not read in full. A retrieval layer grabs passages and drops them into a shared context window next to the user's question, competing for space with other sources. Because recall fades deep in a crowded window, a page that front-loads the answer, stays tightly chunked, and cuts filler is far likelier to get quoted than a long, rambling one. Our AI SEO work structures pages so the parts that matter survive retrieval and land inside the slice of context an assistant actually reads and cites.
Get AI SEO that moves the needle
We turn terms like this into ranked pages and qualified pipeline. Start with a free Initial SEO Strategy.