Why does AI forget what I said earlier in a chat?
An AI can only read a limited amount of text at once, called its context window. In a long chat, early details can fall outside it or get less attention, so the assistant seems to forget. Fix it by restating key facts, asking for a summary and starting a fresh chat, or using memory and project files.
What is happening when an AI forgets
An AI assistant does not remember a chat the way you do. Each time you send a message, the assistant is given the conversation so far and reads it again from the top. The amount it can read in one go has a ceiling. That ceiling is called the context window.
Anthropic describes the context window as Claude's working memory: the amount of information it can work with in a single chat. On paid Claude plans the newest models support up to 1 million tokens, and others support 500 thousand or 200 thousand. A token is a small piece of a word, so these are very large numbers, but long chats with files and pasted text can still reach them.
Part of the window is always kept free for the answer, so the longest possible conversation is a little shorter than the full window.
Why it can forget before the window is full
Running out of room is one reason. There are others you can notice in practice. Your chat may be very long and full of side topics, so the one fact you care about is buried. A fact you mentioned once, long ago, competes with everything said since. And when the chat is summarized to save room, small details can be dropped.
Anthropic says that for users with code execution enabled, Claude can summarize earlier messages when a conversation nears the limit, while the full chat history stays saved. That helps long chats continue, but a summary is shorter than the original, so it is not a perfect copy.
What to do about it
- Restate the key facts. One short sentence that repeats what matters is often enough to bring the assistant back on track.
- Ask for a summary and start fresh. Say: summarize the decisions, the current state and the next steps in under 200 words. Open a new chat, paste it, and continue. The new chat has plenty of room and none of the clutter.
- Keep one topic per chat. Starting a new chat for each task keeps the context clean and also uses less of your plan's allowance.
- Use a project. Both ChatGPT and Claude have projects where you can keep files and instructions that every chat in the project can use. Anthropic suggests projects for working with larger amounts of information.
- Turn on memory if you want facts to carry between chats. See our page on memory across chats for how each assistant handles it.
A good summary prompt
Copy this when a chat gets long: List what we decided, what is finished, what is still open, and any names, numbers or constraints I gave you. Keep it short and exact. I will paste it into a new chat.
Read the result before you paste it. If something important is missing, add it. You are the only one who knows what matters.
Where a node helps, honestly
A node does not make an assistant's context window bigger. Every model still has its own limit.
What it can do is hold the important facts outside the chat. Agents that join a node share one thread and one memory, and you choose which files and notes go in. A fresh session reads the notes instead of a long replayed history, so you stop re-explaining. The node runs on your own computer, and each step is written in one thread you can read.
Do it with a NodePress node
- Create a node on your computer and name it for the project.
- Paste the one line into each assistant you use so they share one memory.
- Ask the agent to save key decisions, names and numbers to the node memory as you go.
- Start new chats often, and tell each new agent to read the node notes first.
For a short, one-off question, a node is more than you need. Restating the key fact in the chat is faster. A node also does not fix an assistant that is simply wrong about something.
Questions people ask
Why does ChatGPT forget what I said earlier?
It only reads a limited amount of text at once, and in long chats early details can fall outside that or be diluted. Restate the key facts or start a new chat with a summary.
Why does Claude lose context?
Claude has a context window, its working memory for one chat. Anthropic says it can summarize earlier messages in long chats, so small details may be lost.
How do I stop an AI from losing context?
Keep one topic per chat, save key facts in a project or memory, and ask for a summary before starting a fresh chat.
Is a bigger context window the same as memory?
No. The context window is what the assistant can read right now in one chat. Memory features save facts to use in later chats.
Does starting a new chat help?
Yes, if you bring a short summary with you. It removes clutter and also uses less of your plan allowance than a very long chat.
Related
- Do ChatGPT and Claude remember across chats? How to control it
- Context window limits explained: what fills it and what to do
- How to hand off work between AI sessions and models
- How to give AI agents shared memory
- How to get AI agents talking to each other
- How to get AI agents talking to each other