In lesson 2.5 you met two ideas: the context window, ChatGPT's working space for one chat, and memory, what it carries from chat to chat. We left the mechanics for later. Later is now.
This one pays off fast. Once you can picture what fills the window, how memory feeds it, and why a long chat goes fuzzy, you stop fighting ChatGPT and start steering it. The reward: sharper answers, fewer do-overs, and chats that don't fall apart at the two-hour mark.
First, a token
ChatGPT doesn't read letters or whole words. It reads tokens: small chunks of text. OpenAI's own rule of thumb: a token is about four characters, or roughly three-quarters of an English word, so 100 tokens is about 75 words. Other languages, Dutch included, often take a few more tokens for the same text.
Why care? Because tokens are the unit everything is counted in. The window's size is in tokens. Long inputs take longer to process. And on plans with usage-based allowances, like Work and Codex, bigger jobs use more. So "how many tokens is this?" really means "how much space and time will it take?"
What the window actually holds
Picture your desk. The context window is the desk surface: everything ChatGPT can look at while it answers. Not your whole archive, just what is on the desk right now, in this chat.
Four things share that surface:
- The files you attached. Uploads, pasted text, results pulled in from apps.
- The conversation so far. Every message, yours and ChatGPT's.
- What memory brings in. Your custom instructions and any saved memories or past-chat details ChatGPT judges relevant.
- The reply being written. Space is always kept for the answer.
How big is the desk? It depends on your plan and your setting. On Plus, Instant has about 54,000 tokens, while the thinking models get about 256,000, roughly 320 pages of text. Pro goes to 128,000 and 400,000. OpenAI doesn't publish one number for Business, and it varies by model, so the practical rule is the same everywhere: thinking levels give you a much bigger desk than Instant. It still isn't endless.
Window versus memory
People mix these up, and it causes real confusion. The window is the desk; memory is a filing cabinet next to it.
The desk is wiped clean when you start a new chat. The filing cabinet isn't: saved memories, chat history, and your custom instructions sit there between chats, and ChatGPT pulls the relevant bits onto the desk when it answers. That is why a brand-new chat can still "know" your role.
Two consequences. First, a new chat clears the desk but not the cabinet, so starting fresh doesn't lose who you are. Second, whatever memory brings in also takes desk space, and if it's out of date, it quietly steers the answer. Lesson 2.5's habit, checking the memory summary now and then, keeps the cabinet honest.
Why long chats drift
The desk is big, but it fills. As a conversation grows, early details get harder for the model to keep track of, and in the end ChatGPT simply stops: "You've reached the maximum length for this conversation, but you can keep talking by starting a new chat."
Well before that, you may feel the drift:
- It repeats a point it already made.
- It forgets an instruction you gave near the top.
- It gets blander than it was an hour ago.
- It contradicts something you agreed earlier.
There's a subtler version too. Even below the limit, a cluttered window makes it harder to find the part that matters, the way a crowded desk buries the one page you need. What is on the desk matters as much as how much fits.
When to start fresh, and how to carry work over
The fix is a habit: one topic per chat, and start fresh when the topic changes or the answers start slipping. A new chat is a clean desk.
You don't have to lose the thread. Hand the work over on purpose:
Before we stop, write a short handoff I can paste into a new chat.
Cover: what we've decided, the current draft, and the open questions.
Keep it tight.
Paste that into a fresh chat and carry on. For work you return to often, put it in a project (lesson 4.3), so the files and instructions are waiting next time.
Keep the window working for you
- Feed it lean material. Plain text and Markdown carry the same words as a heavy PDF or styled Word file for far less space.
- Bring in only what the task needs. Every file, app result, and search adds to the desk. Mention the one app you need instead of pulling from five.
- Pick the level for the job. Long documents and big analyses fit better on a thinking level; quick tasks don't need the extra room.
- Keep memory accurate. Delete memories that are wrong or outdated, or use a Temporary Chat for one-offs that shouldn't feed it.
The rule of thumb: a clean, focused chat beats a giant sprawling one every time.
Where this goes next
You now hold the levers that decide how well any long piece of work goes: what you put on the desk, what the cabinet brings in, and when you clear it. Next in Part 4 we start reusing your best work: skills and plugins package a job so ChatGPT does it the same way every time, and projects give ongoing work a home.
Try it
Take a long chat you already have, or push a fresh one on a real task until it starts to slip. Then test the handoff: paste the summary prompt above, open a new chat, paste the summary in, and ask your next question there.
Last, open Settings, Personalization, Memory and read what ChatGPT keeps about you. Anything that would mislead a fresh chat? Fix it. Clean desk, honest cabinet: that's the whole lesson.
(General exercise. Role and industry versions come once accounts are in.)