How Does AI Remember Conversations (and Why Does It Forget)?

You tell a chatbot you're vegetarian, and a week later it suggests a meat-free recipe. Yet in another long chat, it forgets something you said ten minutes ago. How does AI remember conversations, and why does it forget? This guide explains AI memory in plain language, with steps to control what it keeps.

Short-term memory: the context window

Here's something surprising: an AI model has no memory between messages. When you send message number ten, the app sends the model the entire conversation so far (messages one to ten) together with your new question. The model reads it all again and replies. That's how it "remembers" what you said earlier.

Analogy: imagine a helper who forgets everything the moment they stop talking, but you hand them the full written transcript before each reply. As long as the transcript fits on their desk, they seem to remember perfectly.

The size of that desk is the context window, measured in tokens (small chunks of text; see how AI chatbots work). Modern models have large windows, some holding hundreds of thousands of tokens or more, but every window has a limit.

Why AI forgets in long chats

• The window fills up. In very long chats, the oldest messages may be dropped or summarized to make room.
• Attention spreads thin. Even inside the window, models can pay less attention to details buried in the middle of a long conversation.
• Topic switching. Jumping between unrelated tasks in one chat makes it easier to mix things up.
• Files take space. Uploaded PDFs, images and videos use up the same window, as explained in how AI watches videos.

Long-term memory: saved memories and chat history

To remember you across separate chats, assistants add a memory layer on top of the model:

• Saved memories: short notes such as "Prefers answers in bullet points" or "Is learning Arabic". When you start a new chat, relevant notes are added to the context behind the scenes.
• Chat history reference: some assistants can look back at past conversations for relevant details.
• Custom instructions: text you write yourself, such as "I'm a beginner; explain simply", included in every chat.

Think of it as a notebook the assistant keeps about you. It's the notebook that remembers, not the AI's "brain". The model's trained weights don't change, as explained in how AI learns.

Memory in popular assistants

Names, defaults and availability change often and differ by plan, country and personal vs. work accounts.

For project-style memory with your own files, see Gemini Notebook vs Claude Projects.

"How does ChatGPT know my name?"

This is a very common question. The usual answers are:

1. You mentioned it in an earlier chat and memory saved it.
2. Your account profile includes your name.
3. You connected an app (like email or calendar) that shared it.
4. You told it in the current conversation, and it's still in the context window.

Check your saved memories: you'll usually find the source there. You can delete individual memories or clear them all.

How to control what AI remembers

1. Review saved memories every few months and delete anything outdated or private.
2. Use temporary or incognito chats for one-off questions you don't want stored.
3. Turn memory off entirely if you prefer each chat to start fresh.
4. Don't share sensitive data: passwords, ID numbers, bank details or health information about others.
5. Check training settings separately: memory and model training are different controls.

Memory vs. learning: an important difference

People often mix these up. Memory means saved notes and history fed back into future chats. Learning means changing the model through training. Your chats may be used to train future models depending on your settings, but that's not how the assistant remembers your name tomorrow. Turning off memory doesn't automatically turn off training use, and vice versa.

Can memory be wrong?

Yes. A memory might be saved from a misunderstanding ("User has three children" when you were talking about your neighbor), or become outdated. Wrong memories can quietly affect answers, so if replies seem oddly personalized, check what's saved. And remember that a chatbot can still hallucinate details it never actually stored.

How long does AI remember?

It depends on which kind of memory you mean. Short-term memory lasts only as long as the conversation fits in the context window. Saved memories usually last until you delete them or turn the feature off. Chat history is kept according to each company's retention policy and your settings, and deleted chats are generally removed from your history and later from the company's systems after a set period described in its privacy policy. Temporary chats are designed not to be saved to your history or used for memory.

If you want an assistant to remember something specific, you can usually just tell it: "Remember that I prefer short answers."

Summary

AI remembers within a chat by rereading the conversation each time, limited by its context window, which is why very long chats can lose details. Across chats, memory features save notes and past-chat information and add them back in, without changing the model itself. Review and delete memories regularly, use temporary chats for private questions, and never share sensitive details.

Frequently asked questions