Context windows explained simply
The thing everyone notices but few people understand
If you’ve had a long conversation with an AI and noticed it seemed to “forget” something you mentioned earlier, you’ve encountered the context window limit.
Understanding context windows explains a lot of otherwise confusing AI behaviour — and knowing how to work with them makes you a noticeably more effective user.
What a context window actually is
A context window is the total amount of text an AI model can hold in its working memory at one time. Everything in the conversation — your messages, the AI’s responses, any documents you’ve shared — counts against this limit.
Think of it like a whiteboard. The AI can only work with what’s written on the whiteboard. When the whiteboard gets full, older content gets erased to make room for new content. The AI can no longer “see” what was erased.
Imagine having a conversation with someone who can only hold the last 50 pages of notes in front of them at once. As the conversation grows past 50 pages, they start losing access to earlier parts. They're not forgetting — those pages just fell off the edge of their desk.
Context windows in numbers
Context window sizes are measured in “tokens” — roughly 0.75 words per token, so 1,000 tokens is approximately 750 words.
Current context windows (as of 2026): Claude handles up to 200,000 tokens (about 150,000 words — a full novel). GPT-4o handles 128,000 tokens. Gemini 1.5 Pro handles up to 1 million tokens.
In practice, for most everyday conversations, you’ll never hit these limits. They become relevant when you’re working with very long documents or very long conversations.
How this affects you in practice
Very long documents: If you paste an entire book and ask questions about it, the AI is working near its limits. Questions about early parts of the document may get worse answers than questions about recent parts.
Long conversations: A conversation that has been going for hours with lots of back-and-forth may start producing answers that ignore context established early on.
Inconsistency in long tasks: If you notice an AI contradicting itself across a long conversation, context window limitations are often the cause.
How to work with context windows
Start fresh conversations for new topics rather than extending a single conversation indefinitely. When working with a long document, ask specific questions about specific sections rather than asking broad questions about the whole thing. If something important from earlier seems to have been forgotten, restate it: “As I mentioned earlier, I’m writing for a non-technical audience — please keep that in mind.”
Start a conversation and mention a specific detail early on — for example, your name, or a specific constraint on the task. Have a long back-and-forth about something else (10+ exchanges). Then ask a question where that early detail matters. Notice whether the AI still applies it. This gives you a practical feel for how context persistence works in real conversations.