What is the AI Context Window?
The fuller breakdown of the same idea — the desk, the tokens, and why a bigger window isn't the fix you'd expect. A step up in detail from this page.
Read the explainer →You're a few messages into a good back-and-forth with ChatGPT or Claude. It's actually helping. Then out of nowhere it acts like it never read the first half of the conversation — it drops a detail you gave it, asks something you already answered, or contradicts itself from five minutes ago. You didn't do anything wrong, and the tool isn't broken. You hit its context window.
Here's the good news: you don't need to write a single line of code to understand this — or to work around it. The context window is just the limit on how much your AI can hold in its head at one time. Once you can picture where that limit sits, the "forgetting" stops being a mystery and becomes something you can plan around. This is the plain-English version.
An AI has two very different kinds of memory. There's everything it learned during training — broad general knowledge, like the years of experience a tradesperson walks onto a job with. And there's its short-term memory for the conversation you're having right this minute: what you've typed, what it's replied, and anything you pasted in. The context window is that second kind — the short-term memory — and it has a fixed size.
Picture a whiteboard. Everything the AI needs to answer you has to fit on the whiteboard at the same time: your question, the chat so far, the document you dropped in. If it's on the board, the AI can use it. If the board filled up and something got wiped to make room, that something is simply gone as far as this conversation goes. It isn't tucked in a drawer the AI can reopen — it's erased.
That's the entire idea. A whiteboard that only holds so much, and a conversation that keeps writing on it.
Every time you send a message, the AI re-reads the whole conversation from the top to write its next reply. It doesn't truly remember the chat — it re-reads it, start to finish, every single turn. The longer the conversation gets, the more it has to cram onto that whiteboard each time. Eventually the board fills up.
When it's full, the oldest material gets wiped to make room for what's new. Usually that's a detail you mentioned way back at the start — the budget, the customer's name, the one rule that mattered most. From the AI's point of view, you never said it, because it's no longer on the board. That isn't the AI ignoring you. It genuinely can't see it anymore.
These three symptoms almost always mean you've filled the window — not that the AI got dumber:
Once you recognize the pattern, the fix is usually obvious — and none of it takes a technical background.
The single most reliable fix when an AI gets stuck, repetitive, or forgetful is the simplest one: open a brand-new conversation and paste back in only what matters. You're wiping the whiteboard clean and writing just the essentials on it. Nine times out of ten, the "broken" AI was only a crowded board.
A good rule of thumb: if you've moved on to a genuinely different task, start fresh. If you're three messages deep in repeating yourself, start fresh. It costs you nothing, and it's faster than wrestling a cluttered chat back into shape.
"Most of the time, the AI didn't get dumber. The whiteboard just got full. Wipe it, write back only what matters, and it 'remembers' again."
The habits above cover everyday use. But if you find yourself feeding the AI the same big pile of information over and over — your pricing, your policies, your past jobs — retyping or re-pasting it every time gets old fast, and it never all fits on the board anyway. That's the point where you stop fighting the whiteboard and set up a system that hands the AI only the relevant page for each question, automatically. (The technical name is retrieval, or RAG — but you don't need to know that to benefit from it.)
This is also why some AI setups feel like they genuinely "know your business" and others don't. It usually isn't a smarter model. It's a system built around the context window instead of against it — so the right information lands on the board at the right moment, without you babysitting it.
The context window is the whiteboard. The real skill is deciding what goes on it — and there are exactly four homes for any fact your AI needs. The 51-page plain-English playbook lays out that four-slot system, with a simple rule for routing any fact to the right home, plus a companion checklist you paste into your own AI to get a custom blueprint for your business in about twenty minutes. No coding required.
Real cell, real human, no bot.
← yes, an AI answers. That's the demo.
For longer questions and file attachments.
Fond du Lac, WI · serving the Midwest and remote installs nationwide.