What is a context window?
A context window is the amount of information an AI model can work with in one request. It is measured in tokens: pieces of text rather than whole documents or a fixed number of words. Instructions, conversation history, supplied material, and the model’s output compete for a limited space.
Context is different from memory and training
A document included in a conversation can inform that response without becoming part of the model’s training. An app may also save information between conversations. That product feature is different from the context available to the model for a particular request.
Why does it matter for long documents?
A file upload does not necessarily mean every page reaches the model at once. An application may select passages, summarize material, or omit earlier messages. Before relying on an answer about a long record, check what material the tool actually used.
“Even with large context windows, practical strategies are needed when dealing with voluminous case files or document productions.”
Jonathan Nessler, Infinite Counsel, p. 105.
Working within the limit
State a specific question, provide the relevant passages, and ask for references back to the supplied text. For a long document, work through sections and check the final synthesis against the source. A larger context window offers more space; it does not guarantee correct reasoning.
OpenAI documentation: conversation state and context windows
