An artificial intelligence assistant’s history fills up without anyone thinking about it. A few months of use are enough to leave dozens of saved conversations, a memory holding details about you and, sometimes, an app or two connected for a trial and then forgotten. None of this is a mistake. But each item stays there until you decide otherwise.
Working out how to delete AI chat history and memory needs no technical skills. It needs a sequence of simple steps, taken in the right order, and the awareness that history, memory, training and connected apps are separate controls: removing one does not remove the others.
This guide is the hands-on side of the recommendation to check your AI assistant’s data settings. It follows four phases: prevention, so that the assistant keeps less from the start; detection, to see what it already has; response, to remove what you do not need; recovery, to use your rights and put things right if something has gone wrong. The controls you will find in the settings, with their strengths and limits, are described in the post on AI assistant privacy controls.
Before you start: why order matters
Not everything an assistant keeps carries the same weight. It makes sense to start with what can cause the most trouble and leave for last what is useful to you.
| Priority | What | Why |
|---|---|---|
| 1 | Shared links still live | Anyone with the link can read the conversation |
| 2 | Connected apps with permission to send or edit | They can act in your name on email, calendar or files |
| 3 | Memory holding details about you and other people | It travels from one conversation to the next and may be wrong |
| 4 | Chats with uploaded documents or other people’s data | They contain more than needs to be kept |
| 5 | General history | Useful to pick up again, but to be kept only as long as you choose |
A useful note: if you use the same assistant on phone, tablet and computer, the settings usually apply to the account rather than to each device. It is easiest to tidy up on the device where you can read most comfortably, often the computer, and then check the others.
1. Prevention: the assistant keeps only what you decide
Choose how long the history lasts
Many services let you set conversations to delete automatically after a period you choose, or not to save them at all. The available periods vary from service to service and are given in the help pages. Choose a period that matches how long you really need to pick a chat up again.
Decide what the memory may remember
If the service offers memory, you can usually leave it on, turn it off or, in some cases, ask it to show you new memories before saving them. Memory on is handy for people who use the assistant for work; for occasional questions, it is often not needed.
Choose on training
On many services a switch decides whether your conversations may help improve the model. Leaving it on, deliberately, is as legitimate a choice as turning it off. Bear in mind, though, that it usually works going forward: a large provider states that data already included in training that has started stays there. And on some services a thumbs up or down rating can share the whole conversation even with training off. Where the setting is today in the main services is explained in the resource on how to opt out of AI training.
Use the temporary chat for sensitive questions
A temporary chat stays out of your history and, on many services, neither updates the memory nor is used for training. It is still kept for a short period, given in the service’s help pages, and does not make you anonymous: it is a way of not building up a record, not of disappearing.
Connect only what you use, with minimum permissions
Before connecting email, calendar or file storage, ask yourself what the assistant needs to do: read, write or send. If reading is enough, grant nothing more.
2. Detection: knowing what is there
Scroll through the history
Open the list of conversations and look at it as an archive. Count them, look for those with uploaded documents, those about other people and those you would never reopen.
Reread the memory
If the service shows what it remembers about you, read it line by line. Look for three things: wrong details, details about other people and details you do not remember giving. The signs to watch for most closely are collected in the post on the signs your AI assistant remembers too much.
Look at the list of connected apps
For each one, check since when it has been connected and which permissions it has. If you cannot remember why you connected it, it almost always needs to go.
Some services gather on one page all the conversations you have shared with a link. If there is one, open it: a link sent once to a friend may still be live.
Check devices and sign-ins
If the service shows where the account is signed in, make sure there are no devices you do not recognise. An unknown sign-in changes the priorities: in that case, go straight to the recovery phase.
3. Response: removing what you do not need
Remove the links you no longer need, from the list of links or from the conversation itself. If you had posted the link somewhere, delete that post or message too.
Disconnect the apps you do not use
Remove unnecessary connections and reduce the permissions of those that remain. If the connected app is a service you reach with a “Sign in with…” button, the steps are the same as in the guide on disconnecting apps from your social account.
Delete memories first, then chats
Delete from the memory any wrong details, details about other people and anything you do not want kept. Do this before deleting the chats, and separately: on at least one of the main services, according to its documentation, deleting a conversation does not necessarily delete what the assistant remembered from it.
Export first, if you need to
If you want to keep a record of what you have built with the assistant, many services let you download a copy of your data. The file contains everything you have written: save it somewhere protected, do not leave it in your downloads folder, and delete it when you no longer need it.
Delete the conversations
Delete chats with documents or other people’s data, and those you no longer need. If a chat contains something you want to keep, copy the text somewhere of your own before deleting it. Remember two limits: on many services a deleted chat is removed from the provider’s systems after a certain time, stated in the help pages; and on at least one of the main services, conversations already reviewed by human reviewers are kept even when you delete your activity.
4. Recovery: rights and remedies
If a service lacks the switch you want, object
In the UK and the EU, the UK GDPR and the GDPR include the right to object (Article 21): when a service uses your data on the basis of its own legitimate interests, meaning a purpose of its own that does not require your consent, you can ask it to stop. The European Data Protection Board (EDPB), in its Opinion 28/2024, notes that for AI models this right applies and should be ensured. As a rule the service must stop, unless it can show compelling legitimate grounds that override your interests.
Ask for erasure, if needed
You can ask for erasure of data that is no longer needed or is being used unlawfully. Organisations must respond to rights requests within one calendar month, extendable by up to two months for complex ones, as long as they tell you why: the Information Commissioner’s Office (ICO) sets this out for objections, and the GDPR applies the same timescale to erasure. A request needs no special form, and an email will do. If you are not satisfied, you can raise a concern with the ICO; in the EU, you can complain to your national data protection authority. Keep a copy of your request and the date.
If someone has used your account
Change the password from a device you trust, use one you do not use anywhere else and turn on two-step verification. Sign out of sessions on devices you do not recognise, and check the connected apps and the memory, which the intruder may have changed.
Revoke the link, delete the post or message where you had published it and, if the conversation contained other people’s data, let them know simply: they can then be more careful about unusual messages in the following days. If the conversation included your address, phone number or details of an ongoing matter, treat any related-looking request for payment with caution and check it through a channel you already know.
If you pasted work data into a personal assistant
Tell your organisation’s security contact or IT team. It is not an admission of guilt: it is the information that lets them judge whether anything needs to be done.
A plan in three sessions
You do not need to do everything at once. Three short sessions are enough.
| Session | When | What to do | Time |
|---|---|---|---|
| 1 | Today | Revoke shared links, disconnect apps you do not use, reread the memory | 10 minutes |
| 2 | Within a week | Delete chats with documents or other people’s data, choose how long the history lasts, decide on training | 15 minutes |
| 3 | In a year’s time, or at the next announced change | Repeat the check on all your personal assistants | 10 minutes |
Common mistakes when putting things right
- Deleting the account out of fear. It removes everything, including what was useful, and on many services it does not touch what has already been used for training. It can help, but it is not needed to tidy up.
- Deleting chats and leaving the memory. Memories can outlive the conversations they came from.
- Revoking the link and leaving the post. The link no longer works, but the text that went with it may already say a lot.
- Turning training off and thinking you have deleted. They are two different things: the first is about the future, the second about the past.
- Trusting an email that invites you to “review your data”. To reach the settings, always start from the app or the web address you know, never from a link you have been sent.
- Tidying up only one assistant. Many people use two or three, sometimes without noticing, inside other programs.
Operational checklist
Prevention
- ☐ I have chosen how long to keep my history
- ☐ I have decided whether to keep memory on
- ☐ I have made a deliberate choice on training
Detection
- ☐ I have reread the memory and the list of connected apps
- ☐ I have checked shared links and the devices signed in to my account
Response
- ☐ I have revoked links and disconnected the apps I do not need
- ☐ I have deleted unwanted memories and then the chats
Recovery
- ☐ I know how to object to a service or ask for erasure
- ☐ I know what to do if someone uses my account
Mini-scenario
Lucy has two young children and a husband she shares the family laptop with. She uses an assistant for the week’s meal plans, her emails as a freelancer and, once, to rewrite a complaint to her car insurer. One evening her husband opens the assistant for a search and is offered “the next step in your complaint”. Lucy realises she does not know what that account contains.
She follows the priority table. She finds no shared links. Among the connected apps is her calendar, with permission to change appointments: she reduces it to read-only. In the memory she finds fifteen lines, including the name of the insurer and the car’s registration number: she deletes them. Then she deletes the complaint chat, after copying the final text into a document of her own. She chooses a shorter auto-delete period and decides to leave training on for meal plans, but to use the temporary chat for personal matters.
Finally, she and her husband agree on a rule: on the family laptop, each of them uses their own account, or a temporary chat. A year later the check takes five minutes: Lucy already knows where to look.
How this connects to the Cyber Welfare Framework
| Pillar | What this content contributes |
|---|---|
| Skills | Knowing how to delete memories and chats, revoke a link, reduce a connected app’s permissions and write an objection request |
| Awareness | Understanding that memory and history are separate, and that turning training off does not erase the past |
| Secure Behaviour | Tidying up in the right order and repeating the check every year or after an announced change |
Reference level: FL3 — Autonomous. This is the level at which tidying up is no longer an occasional fix but a routine that covers all your assistants and includes your rights when the settings are not enough. To measure your other digital habits, take the digital resilience self-assessment.
Conclusion
Deleting AI chat history and memory is one of the simplest privacy habits there is, if you do it in the right order: shared links and connected apps first, then the memory, then the chats, and finally the settings that decide what builds up from tomorrow.
The result is an assistant that still helps you, but keeps and remembers only what you have decided.
Related content
- Check your AI assistant’s data settings — the recommendation this guide builds on
- AI assistant privacy controls — the controls available, with their strengths and limits
- Signs your AI assistant remembers too much — the indicators to check
- What sharing with an AI assistant can lead to — what a tidy-up helps you avoid
- Disconnecting apps from your social account — the related topic of services connected through “Sign in with…”
Related resources
Short reads from the Resources section, for anyone who wants to focus on a single aspect:
- Opt Out of AI Training: Check Your Chatbot Settings
- AI Chatbot Privacy: Where Your Conversations End Up
Start with the first step: the Cyber Welfare Programme guides you free of charge, one recommendation at a time.




