Does ChatGPT Train on My Data? ChatGPT, Gemini & Claude

Does ChatGPT Train on My Data? ChatGPT, Gemini & Claude

Emperola Editorial · · Updated · 5 min read

TL;DR

  • Consumer ChatGPT, Gemini and Copilot can train on your chats by default; Claude does not unless you opt in.
  • Business and API plans default to no training — the plan decides, not a setting.
  • Turn it off in ChatGPT Data controls and by disabling Gemini Apps Activity.
  • Temporary chats are not saved and not used for training — the fastest habit to adopt.

The short answer: on consumer accounts, ChatGPT, Gemini and Copilot can use your conversations to improve their models, while Claude does not do it unless you opt in. On business and API plans the default is the opposite — your data is not used for training.

This page compares the four assistants by what they collect, where the switch is, and what changes on a work account. It is written for people who want to know what to turn off, not for lawyers.

Comparison at a Glance

Assistant Trains on chats by default Where to switch it off
ChatGPT (consumer) Yes Settings → Data controls → Improve the model for everyone off
Google Gemini (consumer) Yes, via Gemini Apps Activity Turn off Gemini Apps Activity
Anthropic Claude (consumer) No — opt-in only Settings → Privacy
Microsoft Copilot (consumer) Yes Privacy settings of the account or app

Retention differs by provider and plan, and every provider can change it. Check the current policy page before you assume a number — what follows are the switches that matter.

ChatGPT (OpenAI)

Consumer ChatGPT uses conversations to improve models unless you turn it off. The switch lives in Settings → Data controls → Improve the model for everyone. Turning it off stops future conversations from being used, but does not remove what was already processed.

Two details people miss:

  • Temporary chats are not saved to history and are not used for training — useful for a one-off question you would rather not keep.
  • Deleting a conversation removes it from your history; OpenAI states deleted conversations are removed from its systems within 30 days, except where they must be kept for legal or abuse reasons.

Google Gemini

Gemini is trained from Gemini Apps Activity, which is on by default in consumer accounts. Review your prompts, uploads and generated answers under myactivity.google.com/product/gemini and turn the activity off — the setting also controls whether human reviewers may read your chats.

Google's default auto-delete window for activity is 18 months and can be shortened to 3 months. Gemini in Workspace (Gmail, Docs) follows the Workspace terms instead: Google states Workspace data is not used to train Gemini without permission.

Anthropic Claude

Claude's consumer product is the exception in this comparison: conversations are not used for model training unless you opt in through the privacy settings. That switch is a genuine choice, not a hidden default — but it also means that if you enabled it, you can switch it back off at any time.

Claude Pro and Team plans follow the same privacy toggle; the API and enterprise deployments do not train on your inputs by default.

Microsoft Copilot

Consumer Copilot can use your interactions to improve Microsoft's models. The controls sit in your Microsoft account privacy settings and inside the app, and they differ between the web, Windows and mobile versions — check both places before assuming it is off.

Microsoft 365 Copilot for organisations is covered by the enterprise agreement: prompts and responses in your tenant are not used to train foundation models.

How to Opt Out in Five Minutes

  1. ChatGPT: profile icon → Settings → Data controls → turn Improve the model for everyone off. Then open Manage next to data controls and review the export and deletion options.
  2. Gemini: open myactivity.google.com/product/gemini, turn the activity off and choose an auto-delete window of 3 months.
  3. Claude: Settings → Privacy — leave the training toggle off, and delete old conversations you no longer need.
  4. Copilot: Microsoft account privacy settings plus the in-app privacy panel; on Windows also check Privacy & security → Diagnostics if you use Copilot there.

Do this once per account, not once per device — the settings follow the account.

What About Work Accounts?

Work and API plans are the part of this topic that is actually settled:

  • OpenAI API and ChatGPT Business/Enterprise: data is not used to train models by default, and retention is governed by the agreement.
  • Google Workspace: Workspace data is not used to train Gemini without permission, and administrators control the services that are on.
  • Anthropic API: inputs and outputs are not used for training by default.
  • Microsoft 365 Copilot: prompts and responses inside the tenant are not used to train foundation models.

That does not make work accounts risk-free — the risk shifts from training to access control: who can read a shared chat, how long logs are kept, and what employees paste in. If you are evaluating an assistant for a company, the enterprise ChatGPT guide goes through the data-flow questions worth asking.

If you would rather not depend on switches at all, running a model locally removes the question entirely — see how to run local AI models offline.

FAQ

Does ChatGPT train on free users' chats?

Yes, by default, unless you turn off Improve the model for everyone in Data controls. Business, enterprise and API usage follows a different default, where training is off unless an organisation explicitly opts in.

Does Gemini read my Gmail?

Not for training by default: Google states Workspace data is not used to train Gemini without permission. Consumer Gemini Apps Activity is separate and covers the prompts and uploads you make in the Gemini app itself.

Is Claude training on my data?

Not by default. Anthropic's consumer product trains only if you opt in through the privacy settings; the API and enterprise offerings do not train on your inputs by default.

Does turning off training delete my old chats?

The switch is forward-looking — it stops future use. Old conversations stay in your history and can be deleted separately; deletion removes them from the service, though copies may persist in logs for a limited period.

Can I use these tools without any training at all?

Yes: turn the training switch off on every account, use temporary chats for sensitive questions, and for anything truly private run a local model so the data never leaves the machine. The AI scraping page covers the web side of the same problem.

Share on

Related articles