Explainers·7 min read

ChatGPT, Claude, and Gemini All Train on Your Chats by Default Now. The Off Switch Doesn't Delete Anything

Quick answer ✦As of September 2026, ChatGPT, Claude, and Gemini all default personal accounts to using conversations to improve their models, and all three treat "stop training on my chats" and "delete my chat history" as two separate settings. OpenAI's toggle is Improve the model for everyone, under Settings, Data Controls. Anthropic's is Help Improve Claude, under Settings, Privacy, on by default for Free, Pro, and Max plans since Anthropic changed its consumer terms in 2025. Google's is Keep Activity, inside the Gemini app's settings. Turning any of the three off stops future chats from training the model. None of them erase what's already been said until you take a separate deletion step.

The confusion here is reasonable. A toggle labeled "improve the model" or "help improve Claude" reads like a preference about the product, not a data-sharing decision, and none of the three apps make that distinction obvious at the moment you flip the switch.

Business and API accounts are the exception across the board: Team, Enterprise, Education, and Claude for Work plans, plus every API call to any of the three, default to no training at all. This piece is specifically about what happens on an ordinary, personal account.

ProviderConsumer defaultToggle locationTurning it off also deletes past chats?
ChatGPTTraining onSettings, Data Controls, Improve the model for everyoneNo. New chats get roughly a 30-day retention window before deletion, per OpenAI's own help article.
ClaudeTraining on, since Aug 2025Settings, Privacy, Help Improve ClaudeNo, and safety-flagged chats can still be kept regardless of this setting, per Anthropic's privacy center.
GeminiTraining on, accounts 18+Gemini app, Settings, Keep ActivityNo, you must separately choose "Turn off and delete activity," per Google's Gemini Apps Privacy Hub.

Anthropic's carve-out that undercuts a clean opt-out

Anthropic's policy has a wrinkle worth knowing even after switching Help Improve Claude off: conversations flagged by its automated safety systems can still be used and retained for trust-and-safety purposes regardless of that setting, per Anthropic's own model-training help article. Flagged inputs and outputs can be kept for up to two years, and the classification scores tied to them for up to seven.

5 years

how long Anthropic can retain de-identified chats if Help Improve Claude stays on

per Anthropic's privacy center. Turning it off drops new chats back to the standard 30-day window.

18 months

how long Google keeps your Gemini Apps Activity by default before auto-deleting it

adjustable to 3 months, 36 months, or never, per Google's own activity controls page.

72 hours

minimum time Google keeps every Gemini chat, even with Keep Activity switched off

held briefly for abuse prevention regardless of your setting, per Google's Gemini Apps Privacy Hub.

Oct 8, 2025

deadline Anthropic gave existing Claude.ai users to accept training-by-default terms

per Anthropic's own announcement of the change.

The exact toggle for each one

  1. 1ChatGPT: Settings, Data Controls, turn off Improve the model for everyone. This only affects chats going forward.
  2. 2Claude: Settings, Privacy, turn off Help Improve Claude. Safety-flagged conversations can still be retained regardless.
  3. 3Gemini: open the Gemini app, Settings, Keep Activity, and choose Turn off and delete activity if you also want existing history erased, not just future chats stopped.
  4. 4None of these settings apply retroactively to a model that has already finished training on your older conversations. The toggle only controls what happens next.
Do ChatGPT, Claude, and Gemini train on my conversations by default?

Yes, on personal Free, Plus, Pro, and equivalent tier accounts as of September 2026. Business, Enterprise, Education, and API accounts across all three default to no training.

If I turn off the training toggle, does it delete my past chats?

No. All three treat training and history deletion as separate settings. You need an additional step, like Google's "Turn off and delete activity," to erase what's already stored.

Can Anthropic still use my chats even with Help Improve Claude turned off?

Conversations flagged by Anthropic's automated safety systems can still be used and retained for trust-and-safety purposes regardless of that setting.

Does using the API instead of the consumer app change anything?

Yes. API usage across OpenAI, Anthropic, and Google defaults to no training, separate from the consumer app policies described here.

Is there a way to keep my chat history but stop training?

Yes, on all three. Turning off the training toggle alone keeps existing history intact while opting future chats out of training data.

Try the tools ✦

Free browser tools that never upload your files.

Open Tools