Context window: how much an AI can read at once
2026-11-10
Definition
The context window is the maximum amount of text, measured in tokens, that an AI model can take in and consider at one time, including your instructions, any documents and its own reply.
Why it matters
Everything outside the window is invisible to the model. In a long chat, early messages can drop out or lose influence, and the assistant may seem to forget what you agreed. Large windows let you paste whole contracts or reports, but bigger is not automatically better. Answers can get less precise when the window is stuffed with irrelevant material.
Practical habits help. Start a fresh chat for a new task, restate key rules in your prompt and send only the relevant sections. For large document collections, retrieval-augmented generation finds the right passages so you do not need to fit everything in. Window sizes differ by model and change often, so check current documentation such as the Anthropic docs or OpenAI platform docs. Our guide to writing better prompts for work covers how to keep inputs tight.
A hypothetical example
Picture an HR manager who pastes an entire year of staff survey comments into a chat and asks for themes. The tail end of the data gets weaker treatment. Splitting the comments into batches and merging the summaries gives a more reliable result.
Teaching these habits is part of AI adoption.