You can delete the chat where you told ChatGPT you’re allergic to peanuts, and ChatGPT can still remember you’re allergic to peanuts.

That isn’t a bug. OpenAI says so in its own help pages: saved memories live separately from your chat history, so deleting a conversation doesn’t automatically delete a memory created from it. So, how does AI chatbot memory work? The short version: the AI model itself doesn’t remember you at all. The app around it keeps notes about you in several separate places, and each place has its own delete button. Miss one, and the information stays.

This guide follows that single peanut allergy fact through all four places it can end up, shows you what ChatGPT, Claude and Gemini actually keep, and gives you the exact steps to inspect and wipe each one.

The Stateless Illusion: Why AI Models Don’t Actually “Learn” You

Ask “how does AI chatbot memory work?” and most people picture a digital brain slowly getting to know them. That picture is wrong in an important way.

The model behind a chatbot is stateless. That means it doesn’t change between conversations. When you finish a chat, the model is exactly the same as it was before you started. It hasn’t stored anything.

What feels like memory is a trick of packaging. Before your message reaches the model, the app quietly adds extra text to it: your saved preferences, a summary of who you are, maybe snippets from older chats. The model reads it all fresh each time and answers as if it knew you.

Think of it like a doctor with no memory who reads your file before every appointment. The doctor seems to know you. Really, the file does. Delete the file and the doctor has never met you.

That’s why AI chatbot memory is really a question about files: where they’re kept, what’s in them, and who can delete them.

How Does AI Chatbot Memory Work? The Four Layers

Your peanut allergy can sit in four different places. They behave very differently.

Layer 1: The context window (what the AI sees right now)

The context window is the amount of text a model can read at once, like a desk that holds only so many pages. Everything in your current conversation sits on that desk. Say “I’m allergic to peanuts” early in a chat, and the AI can use it ten messages later.

When the chat ends, the desk is cleared. This is the only layer that forgets on its own. This difference is the core of AI memory vs context window: the context window is short-term and automatic, while memory is something the app deliberately saves.

Layer 2: Chat history (the saved transcript)

Your conversations are stored on the company’s servers so you can scroll back through them. Newer features also let the AI search or reference these old chats. ChatGPT calls this “Reference chat history,” and Claude can search your past conversations.

So the peanut sentence still exists here, word for word, until you delete that chat.

Layer 3: Saved memory or profile (the notes about you)

This is the layer most people mean when they search “how does AI chatbot memory work.” The app pulls out facts it thinks are useful (“user is allergic to peanuts”) and stores them as separate notes. ChatGPT now keeps a memory summary that updates as you chat. Claude builds a memory summary too, and gives each project its own separate memory. Gemini uses what Google calls “personal context” from past chats.

How are these notes stored? Companies don’t publish their exact systems, but a common method uses a vector database. Text is turned into a list of numbers that captures its meaning, so “allergic to peanuts” sits mathematically close to “nut-free recipes.” When you ask for dinner ideas, the system finds the nearest notes and slips them into the context window. That’s how AI chatbots remember things across chats without the model learning anything.

Layer 4: Model training (the one you can’t take back)

Training is the one part of “how does AI chatbot memory work?” where deleting things can’t fully undo them. If your chats are used for training, they help shape future versions of the model for everyone. Your peanut allergy wouldn’t become a note about you. It would become a tiny part of the model’s general patterns.

Deleting a chat later stops it from being used in future training, but it can’t pull data back out of a model that has already been trained.

Layer	What it holds	Clears itself?	How you remove it
Context window	The current conversation	Yes, when the chat ends	Start a new chat
Chat history	Full transcripts	No	Delete the chat
Saved memory	Notes and summaries about you	No	Delete the memory itself
Training	Patterns in future models	No	Opt out before it happens

That’s the honest answer to “how does AI chatbot memory work?” Four systems, four sets of controls.

What AI Assistants Actually Keep: Explicit Facts vs. Inferred Patterns

The other half of “how does AI chatbot memory work?” is what actually gets saved. Saved memories come in two flavours.

Explicit facts are things you told the AI directly, like “Remember that I’m vegetarian.” These are easy to spot and easy to delete.

Inferred patterns are conclusions the AI drew on its own. Ask about baby monitors three times and it may decide you’re a new parent. Mention a city twice and it may decide you live there. These are harder to notice, and they can be wrong.

What do AI chatbots remember about you that’s sensitive? Anything you’ve shared. OpenAI’s help pages warn plainly that sensitive information can end up in memory if you share it with ChatGPT. Health worries, money problems, relationship details and work secrets all count.

Why Deleting a Chat Often Doesn’t Erase What the AI Remembers

Here’s the part most people get wrong. If you take one answer away from “how does AI chatbot memory work?”, make it this: deleting a chat only clears Layer 2.

The note in Layer 3 was copied out of that chat and stored on its own. Delete the original conversation, and the copy stays. OpenAI’s Memory FAQ on how to fully remove remembered information says that to fully delete something, you need to remove every source where it appears: past chats, archived chats, files, the memory summary and any connected apps.

Then there’s the gap between “deleted” and “gone.” Here’s how long each company keeps data after you delete it:

• ChatGPT: deleted chats are removed from your account straight away and permanently deleted from OpenAI’s systems within 30 days. OpenAI may also keep logs of deleted saved memories for up to 30 days for safety and debugging.
• Claude: if you’ve turned off “Help improve Claude,” Anthropic keeps your data for 30 days. If you allowed training, it keeps it for up to five years, though deleted conversations aren’t used for future training.
• Gemini: even with activity saving turned off, Google keeps conversations for up to 72 hours. Chats selected for human review are stored separately from your account and kept for up to three years, even if you delete your activity.

Court cases can override this. A 2025 order in The New York Times’ lawsuit forced OpenAI to keep deleted chats until, OpenAI says, it ended on 26 September 2025.

If you’re in the UK or EU, data protection law (GDPR) also gives you the right to request a copy of your data and ask for it to be erased. Each company has a privacy request process for this.

How to Inspect and Delete Your Memory in ChatGPT, Claude and Gemini

Here’s the practical side of “how does AI chatbot memory work?”: where each app keeps your notes and how to clear them. Menu names change often, so if a label looks slightly different, look for the closest match.

ChatGPT

1. See what it remembers: go to Settings > Personalization > Memory and open the memory summary. You can also simply ask, “What do you remember about me?” The summary doesn’t always show everything, so asking is worth doing too.
2. Edit or delete: remove individual items from the summary or saved memories, or tell ChatGPT to forget something specific.
3. Wipe everything: use “Delete and turn off memory” from the three-dot menu. This deletes what’s remembered but not your past chats, so delete those separately.
4. Turn off chat referencing: switching off “Reference chat history” schedules memories drawn from past chats for deletion within 30 days.
5. Stop training: in Data controls, turn off “Improve the model for everyone.”

Claude

1. See what it remembers: open Settings and view the memory summary. You can edit it directly or ask Claude in a chat to change something.
2. Pause or reset: this is the difference between pausing and resetting AI memory. Pausing keeps the existing memory but stops Claude using it or adding to it. Resetting deletes the stored memory, and if you turn memory back on, Claude starts from scratch. Anthropic’s guide to managing memory in Claude covers both.
3. Check projects: each project has its own memory, so review those separately.
4. Stop training: in privacy settings, turn off “Help improve Claude.”

Gemini

1. Turn off personal context: in the Gemini app, go to Settings > Personal context > Your past chats with Gemini.
2. Delete activity: open Gemini Apps Activity (now called “Keep Activity”) to review and delete past chats.
3. Shorten auto-delete: the default is 18 months. You can change it to 3 or 36 months.
4. Turn off Keep Activity to stop future chats being saved to your account or used to improve Google’s AI. Google’s Gemini Apps Privacy Hub explains what this does and doesn’t cover.

	ChatGPT	Claude	Gemini
View memory	Memory summary in Settings	Memory summary in Settings	Personal context settings
Private chat mode	Temporary Chat	Incognito chat	Temporary Chat
Training control	Improve the model for everyone	Help improve Claude	Keep Activity
Kept after deletion	Up to 30 days	30 days (or up to 5 years if training is on)	72 hours minimum; reviewed chats up to 3 years

Practical Privacy Hygiene: Temporary Chats and Training Opt-Outs

Is turning off AI memory enough to protect privacy? No. It stops new notes being made, but your chat history stays, training settings are separate, and in ChatGPT, turning memory back on later can create new memories from chats still in your history.

A better routine:

• Use a private chat mode for sensitive questions. ChatGPT’s Temporary Chats don’t use or create memories. Claude’s incognito chats (the ghost icon) stay out of your history and memory. Gemini’s Temporary Chats don’t appear in your history or personalise future chats. All three are still kept briefly on the company’s side, so they’re private from the app, not invisible.
• Turn off training in every app you use.
• Check your memory page once a month. Delete anything outdated, wrong or too personal.
• Delete the source, not just the memory. Remove the original chat too.
• Keep identifying details out. Account numbers, full addresses, client names and medical records don’t belong in a consumer chatbot.

You can read our Related article about AI chatbot vs AI Agent

Back to That Peanut Allergy

So how does AI chatbot memory work when you want something gone? Treat it as four separate clean-ups. To erase that one peanut fact completely, you’d delete the memory note, delete the chat it came from, check any project memory, and make sure training was switched off before you ever typed it.

The model never knew you. The files did, and now you know where they are.

FAQ

How does AI chatbot memory work in simple terms?

The AI model doesn’t remember anything between chats. The app saves notes about you (and sometimes your past chats) and adds them to each new conversation behind the scenes, so the AI seems to know you.

How do AI chatbots remember things across chats?

Through saved memories, memory summaries and chat history search. When you start a new chat, the app finds relevant notes and adds them to what the model reads before it answers.

Can I edit what an AI assistant remembers about me?

Yes. ChatGPT, Claude and Gemini all let you view and delete remembered information in settings, and ChatGPT and Claude let you edit or correct it directly or by asking in a chat.

Does deleting a chat delete AI memory?

Not necessarily. A memory created from a chat is stored separately, so you need to delete both the memory and the original chat.

Does AI memory train the model?

No, they’re separate. Memory personalises your own chats. Training uses conversations to improve future models for everyone, and it’s controlled by a different setting in each app.

How does AI chatbot memory work in ChatGPT compared with Gemini?

The idea is the same, but the controls differ. ChatGPT keeps a memory summary you can view and edit. Gemini uses “personal context” from past chats, which you switch off separately from your Keep Activity history.

How long do AI chatbots keep deleted conversations?

ChatGPT deletes them from its systems within 30 days. Claude keeps data for 30 days if you’ve opted out of training, or up to five years if you’ve opted in. Gemini keeps chats for at least 72 hours, and chats picked for human review for up to three years