Can an AI remember what you tell it?
Inside one conversation, yes: the model rereads everything above your last message. Across conversations, only if the app has a memory feature, a separate set of notes about you that it brings back each time. Some apps keep nothing between chats. Before telling one anything personal, check whether you can see, edit and delete what it remembers.
Vincent Legardien3 min read

Inside one conversation, yes. Across conversations, only if the app was built to. Those are two different things, and most of the confusion comes from mixing them up.
Picture it. On Monday night you tell a chatbot about the argument with your sister. On Wednesday you open a new chat and write "she called again". Some apps answer "your sister? how did it go?". Others answer "who called?". Same kind of technology, very different experience.
What the AI sees inside one chat
While a conversation is open, the model rereads everything above your last message each time it answers. That's its working space, called the context window. It isn't memory in the human sense: close the chat, and that space is empty again.
Even inside one long chat, recall isn't perfect. In 2023, researchers at Stanford led by Nelson Liu tested language models on long texts and found they used information best when it sat at the very beginning or the very end, and noticeably worse when it was buried in the middle. So if something you said two hours ago seems to have vanished, it may simply have slipped into that middle.
How an AI remembers you from one day to the next
The model itself doesn't learn from you as you talk. What remembers is a separate layer around it: a set of notes the app keeps about you, and slips back in front of the model when you start a new chat.
ChatGPT added this kind of memory in February 2024. According to TechCrunch at the time, you could ask it to remember something, or let it pick up details on its own; you could see and delete each memory in the settings, or turn the whole thing off. One detail surprised a lot of people: deleting a chat did not delete what had been remembered from it. You had to delete the memory itself. Features change often, so check the current settings of whatever app you use.
Other apps keep nothing at all between sessions. Each conversation starts from zero, which is restful if you want privacy, and tiring if you have to explain who your sister is every single time.
What's worth checking before you tell it anything personal
Three questions, and the answers are usually in the settings or the privacy policy:
- Can you see what it remembers? A memory you can read is a memory you can correct. A hidden one is just a guess about you.
- Can you delete it, and does deleting a chat really delete it? Look for a clear "forget" option, not only a delete button on each conversation.
- Is a temporary or private mode available? For the things you'd rather not leave anywhere, a chat that leaves no trace is the right tool.
The rest is common sense. Passwords, bank details, your address: no app needs them to understand your week.
When memory is the whole point
Remembering isn't a gimmick when what you want is to talk about your life. The value of a friend who asks "so, did your sister call back?" is that you don't start from scratch. It's also why I built Lucy around structured memory: the people in your life, what's going on, what you said last week. You can delete it in one gesture, and nothing is sold or shared. But whichever app you pick, the three questions above apply.
Related questions
Does ChatGPT remember everything I say? Not word for word. Inside a chat it sees the whole conversation; between chats, it only uses what its memory feature kept, if memory is on. You can check and delete those memories in the settings.
Can I make an AI forget something? In apps with memory, usually yes: delete the specific memory, or clear all of it. Deleting the conversation alone may not be enough, so check where the memories are listed.
Does the AI learn from what I tell it? Not in real time: the model doesn't change while you talk. Some companies may use conversations to train future models, and most offer a setting to opt out. That's worth checking once.
Sources
- Liu, N. F., Lin, K., Hewitt, J., et al. (2023). Lost in the Middle: How Language Models Use Long Contexts. Transactions of the Association for Computational Linguistics. arxiv.org
- TechCrunch (13 February 2024). ChatGPT will now remember, and forget, things you tell it to. techcrunch.com
- CNIL and PIPC (2026). Generative AI & privacy: 6 questions to ask before using an AI (poster, French). cnil.fr
Vincent Legardien@legardienv
Founder of Lucy AI. Passionate about building technology that helps people feel better, so real connections have somewhere to grow from.