Independent privacy guide · July 2026

Make AI forget you. Here's every lever that exists.

Every chat you've ever had with an AI told it something: your name, your job, your health worries, your relationships, your company's plans. Some of that is deletable in two clicks. Some needs a formal legal request. And some — honestly — can't be pulled back out. This site maps all three, provider by provider.

Editorial guide, not legal advice. Settings move around — verify paths on official provider pages.

The three layers

Where "AI knowledge" about you actually lives.

01

Your stored conversations

Every chat sits on the provider's servers, tied to your account, until you delete it. This is where names, confessions and company details live verbatim.

Deletable — how →
02

Memory & profiles

Features like ChatGPT Memory distill your chats into standing facts: "user is a nurse in Zurich, has two kids, is planning a divorce." Editable and erasable — if you know where to look.

Deletable — how →
03

Trained model weights

If your conversations were used for training before you opted out, their patterns are diffused into the model. No provider can surgically delete them; regulators accept output suppression instead.

Suppressible — how →

Ten-minute cleanup

The order of operations.

  1. Turn off and clear memory.ChatGPT: Settings → Personalization → Memory → Manage / Clear. This deletes the standing "facts about you" file. Step-by-step →
  2. Stop training on your conversations.ChatGPT: Data Controls → "Improve the model for everyone" off. Claude, Gemini, Copilot and Grok each have an equivalent. All providers →
  3. Delete the chats that matter.Deleted conversations are purged from provider systems within ~30 days. For anything sensitive, delete specifically, don't just archive.
  4. File the formal request.GDPR/CCPA erasure via each provider's privacy portal puts your deletion on the legal record — and reaches data the settings don't. How to apply →
  5. Change how you chat going forward.Temporary chats for sensitive topics, no real names for third parties, and a privacy-first workspace. Private alternatives →

Provider map

Which lever matters at each AI company.

The same privacy words mean different things at different providers. Use this map before you start clicking through dashboards.

ProviderSelf-service controlsFormal request worth filing when...Watch out for
OpenAI / ChatGPTMemory, chat history, training toggle, account deletion, privacy portal.You want erasure, do-not-train confirmation, or suppression of model outputs about you.Memory and chat-history referencing are separate. Clearing one does not automatically clear the other.
Google GeminiGemini Apps Activity, Google account privacy controls, activity deletion.You need deletion beyond visible activity or have sensitive personal data in prompts.Reviewed conversations can be retained separately for longer under Google's disclosed review process.
Anthropic ClaudeConversation deletion, project deletion, privacy center, account controls.You need a written deletion response or business/legal proof of removal.Project knowledge and uploaded files can be a separate cleanup job from ordinary chats.
Microsoft CopilotMicrosoft privacy dashboard, Copilot history and account data controls.Your data crosses Windows, Edge, Microsoft 365, Bing or workplace accounts.Consumer Copilot and work Copilot can be governed by different account policies.
Meta AI / xAI GrokAccount data settings plus AI data sharing and personalization settings.Your social-account activity is part of the AI personalization or training concern.AI settings can sit inside broader social network privacy settings, not only the chatbot UI.

Reality check

What deletion can and cannot achieve.

Can delete

Stored chats, files and account records

These are the cleanest targets: conversations, uploads, workspace files, saved memories, account profile data and billing-adjacent account records where deletion is legally allowed.

Can stop

Future use for training

Training opt-outs are forward-looking. They stop new content from entering improvement pipelines, but they do not rewrite models already trained before the switch was changed.

Can suppress

Outputs about you

For trained-model knowledge, the practical remedy is output suppression: the provider configures systems not to generate personal data about you.

Cannot promise

Instant removal everywhere

Backups, abuse logs, legal holds and reviewed samples can have retention windows. Ask providers to state what remains and the legal basis.

Often missed

Third-party mentions

If someone else pasted your name or details into an AI, your own account deletion will not catch it. Use personal-data removal and output suppression requests.

Best prevention

Change the workflow

Temporary chat, training opt-outs, no real names for other people, and one governed workspace beat repeated cleanup after sensitive prompts.

FAQ

The short answers.

Can I make AI companies delete everything they know about me?

Mostly, yes — with one honest limit. You can delete stored chats and memory, opt out of future training, and file a formal erasure request (GDPR Article 17 / CCPA). What can't be done today is surgically removing information already baked into a trained model's weights; providers handle that by suppressing outputs about you instead.

Where do I start?

Three moves, ten minutes: clear the memory feature, turn off training on your conversations, and delete old chats. Then, if you want formal erasure on the record, file a privacy request — our template page has copy-paste letters.

Does deleting a chat really delete it?

Providers state deleted conversations are removed from their systems within roughly 30 days, unless law requires retention (for example, a 2024 US court order forced OpenAI to preserve some consumer chats for litigation for a period). Deletion stops future use; it is not a time machine for data already used in training.

What about really private things I've asked an AI?

Intimate questions — health, sexuality, relationships — can let a system infer things about you that you never stated. That's special-category data under GDPR and the strongest possible basis for a deletion request. We cover exactly what happens to those inferences on the sensitive-data page.

Is there an AI that doesn't hoover up my life?

No cloud AI is zero-data, but you can choose better structures: temporary/incognito chats, training opt-outs by default, and workspaces with contractual privacy. MultipleChat AI (Swiss-made) keeps conversations inside your private workspace and offers enterprise data isolation and EU/Swiss residency — see our private alternatives page.

More on the full FAQ page.