The context window is everything an AI can see at once in a single conversation: your prompt, the files you pasted, and its own replies. Anything outside that window may as well not exist to the model. For agencies, it is the hidden reason an AI seems to forget your instructions halfway through a long working session.
Also known as context length, maximum context
Think of the context window as a whiteboard the AI works on. It is large but finite, measured in units called tokens. Everything in the conversation competes for the same space: the rules you set, the research you pasted, and the model’s own long answers.
When the whiteboard fills up, the oldest material gets wiped to make room for new input. The model does not warn you. It simply stops being able to see what scrolled off the top, which is often exactly where you put your most important instructions at the start of the chat.
Most complaints that an AI ignored what you told it are really context window problems in disguise.
It explains forgotten instructions. Your brand rules did not get ignored. They scrolled out of the window many messages ago and the model can no longer see them.
It limits what pasting everything achieves. Dropping a 200-page deck into a chat does not mean the model read all of it. Anything past the window is invisible, no matter how confidently the model responds.
It rewards good habits. Restating key rules in long chats, and starting fresh when a session gets bloated, produces more reliable output than one sprawling conversation.
A copywriter sets careful brand voice rules at the start of a long drafting session, then works with the AI for two hours. By the end, the drafts have drifted off-voice. Nothing broke. The session simply outgrew the context window, and the original rules fell off the top of the whiteboard. The fix is a fresh chat with a short, reusable brief that carries the rules forward, which holds the voice steady from the first line.
The workshop covers practical habits for working inside the context window, including when to restate rules and when to start clean.