The Decoder / One term

Does ChatGPT use my data for training?

On a personal ChatGPT account, yes by default: your chats may be used to improve OpenAI's models unless you turn off Improve the model for everyone; business, enterprise and API use are not trained on by default.

By Mstimaj Numbers checked 30 September 2026 Updated 30 September 2026 Level Start Here Tool ChatGPT

Sign in to save your place and mark this done. Free, no card.

A single light switch on a dark wall, its plate lit, with a thin cable running from it into the shadow

Yes, unless you switch it off. Personal accounts on ChatGPT, Claude and Gemini all start with training on, each behind one toggle in settings. Turning it off stops future chats being used; it does not pull back what was already used. Temporary or incognito chats on all three are excluded from training. Paid business plans and the developer APIs are not trained on. The toggle takes ten seconds and the table below shows where it is for each.

Key facts

  • 3vendors, 3 toggles: ChatGPT Data Controls, Claude Privacy Settings, Gemini Apps Activity
  • 30 dayshow long a ChatGPT temporary chat may be kept for safety review, per OpenAI's help page
  • 5 yearshow long Anthropic keeps consumer chats when training is on, versus 30 days when it is off
  • 18 monthsGoogle's default auto-delete for Gemini activity; chats may be reviewed by people
  • OpenAI help pages on data controls and temporary chat (as reported; the pages block automated readers); Anthropic consumer terms update and privacy centre; Google Gemini Apps privacy hub; checked 30 September 2026

How do I turn off training in ChatGPT?

Settings, then Data Controls, then switch off Improve the model for everyone. From that point new conversations are not used for training. Chats are still stored in your history until you delete them, and OpenAI’s help page says a deleted or temporary chat may be kept for up to 30 days for safety review before it is removed. Temporary Chat, the toggle at the top of a new conversation, keeps that conversation out of history, memory and training in one move, which is the right setting for anything about a client.

Training defaults and how to opt out, personal accounts, 30 September 2026
Vendor Default on a personal account Where the switch is Kept how long Private mode
ChatGPT Training on Settings > Data Controls > Improve the model for everyone Until you delete; up to 30 days after Temporary Chat
Claude Choice shown at sign-up, option pre-selected on Privacy Settings > Help improve Claude 5 years if on, 30 days if off Incognito chat
Gemini Training on Gemini Apps Activity 18 months by default, 72 hours if off Turn Apps Activity off
Business plans and APIs, all three Not trained on Not needed Per contract Not needed
How long chats are kept, training on HOW LONG CHATS ARE KEPT, TRAINING ON ChatGPT temporary chat up to 30 days months Gemini, default auto-delete 18 months months Claude, training on 5 years months
Vendor statements. Switching training off shortens Claude's window to 30 days and Gemini's to 72 hours for new chats.

Does Claude train on my conversations?

Since the autumn of 2025 it can, on personal Free, Pro and Max accounts, and the choice was presented to every user with the training option switched on. If you left it on, Anthropic keeps your chats for up to five years and may train on them. If you switch it off, in Privacy Settings under Help improve Claude, retention drops to 30 days and nothing new is used. Incognito chats are never used. Claude for Work, Education, Government and the API are not trained on at all, which is why a business account is the clean answer for client material.

Is Gemini reading my Gmail?

Only when you ask it to, and what it reads in that moment is handled under the same activity setting as your chats. The setting is Gemini Apps Activity. On by default, it saves your conversations, may use them to improve Google’s models, keeps them for 18 months unless you change the auto-delete, and Google says a sample may be reviewed by human evaluators. Turn it off and new chats are not used for training, though Google still keeps them for 72 hours to provide the service. The Gmail integration itself is a separate permission you grant or decline.

Is it safe to upload documents to ChatGPT?

Safe from the model? Yes, the model cannot leak a file on its own. Safe from the vendor’s policy? Only if training is off or you are on a business plan, because an uploaded document is part of the conversation and travels with it. Safe from your own obligations? That depends on the document. A client contract, a patient record, a student’s essay or anyone else’s personal data belongs in a business account with training off, or in a local model that never leaves your machine, or nowhere. The toggle protects you from the vendor. It does not protect you from a confidentiality clause you signed.

Ask what did not make sense

Comments go through our own form and are read before they appear. Links are not allowed; describe it in words.

Comments are read before they appear.

The newsletter

Get the next term.

One email when something goes up. The argument, the film and the term, nothing else. One click to leave.

Have a question now? Ask Mstimaj, she knows the archive.