Context Window
The amount of text an AI model can consider at once, why long chats forget instructions and how to work within the limit.
Context window: the amount of text an AI model can take into account at one time. It usually covers your prompt, any files or text you paste, the conversation so far and, in many models, the answer being written.
In plain English
Think of it as the model’s working memory for one conversation. Anything inside the window can influence the answer. Anything that falls outside it may be forgotten, cut off or ignored. It is measured in tokens, not words.
Why it matters
- Long documents: a big report may not fit, or may fit only partly.
- Long chats: early instructions can slip out of the window, so the AI seems to “forget” them.
- Quality: very long inputs can make a model less reliable at finding a specific detail.
How to work with it
- Put the most important instructions at the start and repeat key rules when the chat gets long.
- Paste only the sections you need instead of an entire document.
- Summarise earlier parts of a long conversation and start a fresh chat with the summary.
- Ask the AI to quote the part of your text it is relying on, to check it found the right detail.
Common confusion
- A bigger window does not guarantee better answers.
- Window sizes differ between models and plans, and change over time.
- The window is separate from stored “memory” features, which some assistants offer.
Related terms
Token, prompt, system prompt, RAG.