What AI Platforms Actually Do With Your Chat Data (And How to Opt Out)

For ChatGPT Free, Plus, and Pro personal accounts, data sharing is on by default. OpenAI's help documentation states plainly that "ChatGPT, for instance, improves by further training on the conversations people have with it, unless you opt out" (OpenAI Help Center). Business tiers — ChatGPT Business, Enterprise, and API access — work the other way around: OpenAI does not train on that data by default. If you're on a personal account and haven't touched your Data Controls, your chats are training material.

Anthropic rewrote its consumer terms in September 2025, and the current policy (last revised July 2026) uses a different default than OpenAI's: Claude only trains on your Free, Pro, or Max conversations "if you choose to allow us to" (Anthropic Privacy Center). That sounds safer, and technically it is — but the trade a user makes for opting in is steep. Standard deleted-chat retention is 30 days. If you allow model training, Anthropic keeps that data in a de-identified form for up to five years, and that five-year window also applies to any conversation you rate with the thumbs-up/down feedback buttons, regardless of your main training setting (Anthropic Privacy Center).

Google's help documentation is unambiguous here: "If you're 18 or over, Keep Activity is on by default" (Google Gemini Apps Help). With Keep Activity on, your prompts, uploads, audio, and shared screens are saved to your Google Account, used to improve Gemini's models, and auto-deleted on an 18-month rolling window unless you change it. A subset of that saved activity is also routed to human reviewers "to help improve Google services," and Google is explicit that reviewed chats are retained for up to three years regardless of what your account-level auto-delete setting says (Google Gemini Apps Privacy Hub).

OpenAI's Data Controls FAQ gives the current path directly, and it differs slightly by device.

Web (signed in): Profile icon → Settings → Data Controls → toggle off "Improve the model for everyone"
Web (signed out): "?" icon (bottom-right) → Settings → toggle off "Improve the model for everyone"
Mobile: Side-bar menu → Profile icon → Data Controls → toggle off "Improve the model for everyone"

The setting is account-wide, not device-specific — turn it off once and it applies everywhere you're signed in (OpenAI Help Center). Existing chat history stays visible to you; it just stops being used for training going forward. You can also start a Temporary Chat for a one-off conversation you don't want saved at all — those are automatically deleted within 30 days and are never used for training (OpenAI Help Center).

Since Claude defaults to opt-in for training, most users don't need to do anything — but if you clicked "allow" during the September 2025 prompt (or later joined and it was pre-selected for you) and want to reverse it, the path is:

Desktop/Browser: Name menu → Settings → Privacy → toggle off "Help Improve Claude"
Mobile: Name menu → Settings → Privacy → toggle off "Help Improve Claude"

Turning it off stops future chats from entering future training runs, but Anthropic is explicit that data already in a training run, or in a model that's already been trained, doesn't get pulled back out (Anthropic Privacy Center).

The Gemini toggle lives in Google's My Activity dashboard, not inside the chat interface itself:

Web: gemini.google.com → menu → Settings & help → Activity (or go directly to myactivity.google.com/product/gemini) → tap "On" → Turn off
Mobile app: Menu → profile icon → Gemini Apps Activity → tap "On" → Turn off

Turning this off stops new conversations from being saved to your account or sampled for human review going forward (Google Gemini Apps Help). It does not retroactively delete chats a human reviewer already looked at — those follow a separate three-year retention clock that isn't tied to your account settings at all.

All three platforms treat feedback — thumbs up, thumbs down, "report a problem" — as a separate consent channel from your main training toggle. On ChatGPT, OpenAI states directly: "Even if you have opted out of training, you can still choose to provide feedback... If you choose to provide feedback, the entire conversation associated with that feedback may be used to train our models" (OpenAI Help Center). Claude does the same thing, and stores that feedback data — plus the full conversation attached to it — for five years, independent of whether your general training toggle is on or off (Anthropic Privacy Center). Gemini's feedback flow works identically: submitting feedback with Keep Activity off still hands Google the last 24 hours of that chat, along with any attached files, for human review (Google Gemini Apps Privacy Hub). If you're trying to stay fully out of training pipelines, don't click the rating buttons.

None of the three platforms delete data the instant you hit delete. ChatGPT removes a conversation from your account immediately but keeps it on OpenAI's backend for up to 30 days, a policy OpenAI confirmed it resumed on September 26, 2025 after a court order tied to New York Times litigation had temporarily forced indefinite retention (OpenAI). Claude follows the same 30-day backend window for standard deletions (Anthropic Privacy Center). Both Anthropic and Google also carve out safety review as an exception that ignores your opt-out entirely: Anthropic states that conversations "flagged by our safety classifiers... may still be used to improve our internal trust and safety models" no matter what your training setting says (Anthropic Privacy Center), and Google says plainly that "even if your Keep Activity setting is off... Google still uses your chats to respond to you and help protect Google, our users, and the public, including with help from human reviewers" (Google Gemini Apps Privacy Hub).

| Platform | Trains on chats by default? | Opt-out path | Retention after opt-out/deletion | Human review? |

|---|---|---|---|---|

| ChatGPT (Free/Plus/Pro) | Yes (source) | Settings → Data Controls → off "Improve the model for everyone" | Deleted chats purged from backend within 30 days (source) | Only for abuse monitoring on Temporary Chats; feedback-rated chats may be reviewed for training (source) |

| Claude (Free/Pro/Max) | No — opt-in (source) | Settings → Privacy → off "Help Improve Claude" | 30 days if opted out; 5 years for data already used in training or tied to feedback (source) | Only for safety-flagged chats and submitted feedback, regardless of your toggle (source) |

| Gemini (Apps) | Yes — Keep Activity on for adults (source) | myactivity.google.com/product/gemini → turn off Keep Activity | 72 hours with Keep Activity off; 18 months (adjustable) if on; 3 years for anything a human reviewer touched (source) | Yes — "a subset of chats," disconnected from your account before review, retained 3 years (source) |

The practical takeaway: opting out reduces what each company keeps, but it doesn't reduce it to zero. If a conversation gets a thumbs-down, gets flagged by a safety filter, or lands in a reviewer's sample, it can outlive your privacy settings by years on every platform covered here.