What is a context window?
Why long AI chats forget what you said at the start, what a context window is, and three habits that help.
Ever had a long chat with an AI and noticed it “forgot” something you said earlier? That’s the context window at work.
The maximum amount of text, measured in tokens, that a model can look at in one go.
A desk with limited space. You can spread out only so many pages at once. When the desk is full, the oldest pages have to come off to make room for new ones.
What fills it up
- 1Instructions — the app's hidden system prompt
- 2Conversation — every message so far, yours and the AI's
- 3Attachments — files, pages or documents you've shared
- 4The reply — the answer being written also takes space
When it runs out
Exactly what happens depends on the app. Some drop the oldest messages, some summarise them, and some ask you to start a new chat.
How big are they?
Context windows are measured in tokens. They range from a few thousand tokens in older models to over a million in some of today’s largest ones. Bigger windows let you share whole documents, but they also cost more to use.
Why it matters
- Long chats drift. Details from early on can get lost or blurry.
- Big documents may not fit. A very long file might need to be split up.
- More context isn’t always better. Burying the key point in lots of text can make answers worse.
Practical tips
- 1New topic, new chat — start fresh when you switch subjects
- 2Restate key facts — repeat important details in long chats
- 3Share only what's needed — the relevant section, not the whole manual
A common misunderstanding
“The AI remembers me.” Within one chat, it sees what’s in the window. Across chats, it remembers nothing, unless the app has a separate memory feature that saves notes about you.
A long customer service thread, with attachments and the policy pages may not fit. When the oldest messages are dropped, the AI can forget what the customer said first and ask for the order number again.