AI chatbots: training, memory and chat history

ChatGPT (OpenAI)

Training on your chatsFuture chats only

Account menu → Settings → Data controls → Improve the model for everyone → Off → Done

Your new chats aren't used to train OpenAI's models. They stay in your chat history. OpenAI says this is on by default for personal Free, Plus and Pro accounts.

In the iPhone and Android apps: sidebar → your profile icon → Data controls → Improve the model for everyone. Signed in, the choice applies on every device; signed out, it's saved in that browser only. Voice mode has its own training control (OpenAI doesn't give its path on this page). If you rate a reply with a thumbs up or down, OpenAI says the whole chat may still be used for training.

Source ↗

Or ask through the Privacy Portal

privacy.openai.com → Make a Privacy Request → Do not train on my content

OpenAI says either this or the setting above is enough; you don't need both. The portal also lets you download your data or delete your account.

The portal is for personal accounts only, not ChatGPT Business, Enterprise or Edu.

Source ↗

Memory

Settings → Personalization → Memory

Shows the memory controls your account has, such as Memory, Reference saved memories and Reference chat history. "Delete and turn off memory" (in the ••• menu of the memory summary) deletes what's in the summary and turns memory off, but not your past chats.

Deleting a chat doesn't always delete a memory made from it, so delete both. OpenAI says it may keep logs of deleted saved memories for up to 30 days. Memory and training are separate settings.

Source ↗

Temporary Chat

New chat → Temporary → Personalized or Unpersonalized

Stays out of your history, doesn't create or update memories, and isn't used for training. OpenAI says it may keep a copy for up to 30 days for safety.

You choose Personalized or Unpersonalized before your first message. If you save a Temporary Chat, it becomes a normal chat.

Source ↗

Delete chats or your account

Account menu → Settings → Account → Delete account → Delete

Deleted chats vanish from your account at once and are deleted from OpenAI's systems within 30 days, unless details that identify you have already been removed or they're kept for security or legal reasons. Deleting your account deletes your data within 30 days, with some kept longer where the law requires or allows.

How long deleted chats are kept is from OpenAI's chat retention page; the account steps are from its account deletion page. Deleting your account doesn't cancel a subscription paid through the Apple App Store; cancel that first. Archiving a chat doesn't delete it.

Source ↗

Google Gemini

Keep activity (saving and training)Still kept 72 hours

gemini.google.com → Activity → On → Turn off (or Turn off and delete activity)

Your future chats aren't saved to your activity and aren't used to train Google's AI models, unless you send feedback. Google says Keep activity is on by default if you're 18 or over.

In the Android app: Menu → your profile picture → Gemini Apps activity → On → Turn off. Even when it's off, Google keeps chats with your account for up to 72 hours to reply and for safety, and still uses them to protect its users and the public, including with human reviewers.

Source ↗

Auto-delete

gemini.google.com → Activity → Deleting activity older than … months (or Choose an auto-delete option) → 3, 18 or 36 months, or Don't auto-delete

Activity is deleted after 18 months unless you change it.

To delete now: the same Activity page → Delete → All time, Last hour, Last day or Custom range.

Source ↗

Memory of past chats

gemini.google.com → Settings and help → Personal context → Memory → Off

Stops Gemini using your past chats to personalise its answers. To remove one thing it remembers, delete the chats that mention it.

Only for personal accounts, aged 18 or over, with Keep activity on.

Source ↗

Voice and Gemini Live recordings

Gemini Apps activity settings → Improve Google services with your audio and Gemini Live videos and screenshares

Google says this is off by default and only works with Keep activity on. If it's on, Google can use your audio, Gemini Live videos and screen shares to improve its services.

Google's page names the setting but we didn't record the steps to reach it (Unverified path).

Source ↗

Human review: Google says trained reviewers read some chats to improve its services, and turning off Keep activity stops that for future chats. Chats already reviewed are disconnected from your account and kept for up to three years, even if you delete your activity.

Turning off Keep activity doesn't change other Google settings such as Web & App Activity or Timeline. Google says that under some privacy laws, such as the UK's and the EU's, you can object to it using your data or ask for wrong details about you to be corrected: below a Gemini reply, More → Report legal issue.

Gemini on an Android phone

Stop "Hey Google" opening Gemini

Gemini app → Menu → Your profile picture → Settings → Talk to Gemini hands-free → "Hey Google" off

Gemini no longer opens when it hears "Hey Google". You can still open the app yourself.

Source ↗

Stop Gemini working on the lock screen

Gemini app → Menu → Your profile picture → Settings → Gemini on lock screen → Use Gemini without unlocking off

You have to unlock the phone to use Gemini.

Source ↗

Take Gemini off the power button

Settings → System → Gestures → Press and hold power button → Another action

Holding the power button no longer opens Gemini.

Google says the steps vary by phone maker.

Source ↗

Switch back to Google Assistant

Gemini app → Menu → Your profile picture → Switch to Google Assistant → Google Assistant → Switch

Switches your phone's assistant from Gemini back to Google Assistant.

Source ↗

Remove the default assistant

Settings → Apps → Default apps → Digital assistant app → Default digital assistant app → None (or another app)

With None, no assistant app is set. Google says deleting the Gemini app on its own doesn't remove Gemini as the default assistant.

Google says the steps vary by phone maker.

Source ↗

Microsoft Copilot (personal Microsoft account)

Training (older app)

Profile icon → Settings → Privacy → Training on conversation activity and Training on voice conversations → Off

Your future chats aren't used to train Microsoft's AI models. Microsoft says they can still be used for the other purposes in its privacy statement, such as safety, security and compliance.

This is the Windows and Mac path. In the older mobile app: menu → profile icon → Account → Privacy.

Source ↗

Memory (new app)

Settings → Personalization → Saved memories → Off

Copilot stops saving new memories. Turning it off doesn't delete the ones already saved: Settings → Personalization → Manage (next to Saved memories) → Delete all memories.

Older app: profile icon → your name → Memory → Personalization and memory; delete everything with Memory → Delete all Memory.

Source ↗

Using your Bing, Edge and MSN activity

Settings → Personalization → One shared experience → Off

Microsoft lists this as the switch that links personalisation in Copilot with Bing, Edge and MSN.

Source ↗

Delete a chat

Chats list → The chat → ...More → Delete

Deletes that chat from your history. To delete all of it, Microsoft points to the activity history on the Microsoft privacy dashboard.

In the older app, history is kept for 18 months unless you delete it.

Source ↗

New app: Microsoft says prompts, replies and files in the Copilot app released on 18 Aug 2026 "aren't used to train foundation models" (the main AI models), so it has no training switch. That covers copilot.com and anyone who has updated the app. The older app has the training switch above.

Microsoft's older-app FAQ says "an opt-out of human review is not available".

Claude (Anthropic)

Training on your chatsFuture training only

Your name → Settings → Privacy → Help Improve our AI models → Off

Anthropic says it won't use new chats and coding sessions for future training, and will stop using stored ones in future training runs. Training that has already started, or models already trained, aren't changed.

The same path works in the mobile app. Chats flagged by Anthropic's safety systems may still be used for safety work. If you rate a reply with a thumbs up or down, Anthropic says it keeps that feedback for 5 years. The pages don't say whether the setting starts on or off.

Source ↗

Memory

Settings → Memory → Generate memory from chats → Pause memory or Reset memory

Pause keeps what Claude remembers but stops it using or adding to it. Reset permanently deletes all memories, including project memories.

Anthropic says memory is on by default for Free, Pro and Max. To stop Claude searching your past chats: Settings → Memory → Search and reference chats (paid plans).

Source ↗

Incognito chat

New chat (outside a project) → Ghost icon, top right

Not saved to your history or memory, and not used for training. Anthropic keeps it for 30 days.

Source ↗

Delete chats or your account

claude.ai → The chat in the sidebar (or Chats and tasks) → ⋮ → Delete

A deleted chat leaves your history at once and Anthropic's systems within 30 days.

iPhone: touch and hold the chat → Delete. Android: open it → ⋮ → Delete. Delete your account: your name → Settings → Account → Delete account (on Pro or Max, cancel the subscription and wait for it to end first).

Source ↗

How long chats are kept: with training allowed, Anthropic says it may keep new or resumed chats, with identifying details removed, for up to 5 years. Chats its systems flag for breaking its rules are kept for up to 2 years.

These are the Free, Pro and Max plans. Anthropic says it doesn't train on its business products by default, such as Claude for Work and its API (the service developers use to build Claude into their own apps).

Grok

Training (Grok app)Future chats only

Settings → Data Controls → Improve the model → Off

Your new chats aren't used to train Grok. Feedback you choose to send may still be used.

On grok.com: Settings → Data → Improve the Model. Grok's FAQ is dated May 2025, so labels may have changed. It doesn't say whether the setting starts on or off. It says that in some countries, but not the UK or EU, you can't opt out if you use Grok without signing in.

Source ↗

Training (Grok inside X)

X settings → Privacy & Safety → Data sharing and personalization → Grok & Third-party Collaborators → Data Sharing → Untick the training option

Stops your public X posts and your Grok chats on X being used to train Grok. X says making your account private also stops your posts being used.

The option reads: "Allow your public data as well as your interactions, inputs, and results with Grok and xAI to be used for training and fine-tuning."

Source ↗

Private Chat

New chat → Ghost icon, top right

Not used for training, and deleted from the company's systems within 30 days.

Source ↗

Delete chats or your account

Settings → Data Controls → Delete All Conversations (or Delete Account)

Grok's FAQ says deletion takes up to 30 days.

Grok inside X: Privacy & Safety → Data sharing and personalization → Grok → Delete Conversation History.

Source ↗

Grok's FAQ now names its maker as SpaceXAI LLC; X's help page still says xAI.

LinkedIn's AI training setting

Data for Generative AI ImprovementFuture training only

LinkedIn settings → Data for Generative AI Improvement → Off

LinkedIn and its related companies, including Microsoft, won't use your LinkedIn data to train AI that writes content from now on. Training already done isn't undone. Private messages aren't used either way.

LinkedIn's help page links straight to the setting (linkedin.com/mypreferences/d/settings/data-for-ai-improvement) but doesn't give the menu path. Feedback you send isn't covered: LinkedIn points to its Data Processing Objection form for that.

Source ↗