What Happens to Your AI Prompts After You Hit Send: ChatGPT, Claude, and Gemini Compared

Illustration of AI chat messages flowing into a server

When you type a prompt into ChatGPT, Claude, or Gemini, the text does not vanish when you close the conversation. On consumer plans, all three keep a copy on their servers, some have human reviewers read a sample of conversations, and some use the text to train future AI models. The good news is that each service gives you controls to reduce this. This article walks through what each provider's own published policy says — verified October 2026 — and which settings actually change what happens to your prompts.

ChatGPT: training is on by default, and the opt-out has one loophole

On personal plans (Free, Plus, and Pro), OpenAI enables model training by default. The control is a setting called "Improve the model for everyone" under Settings, then Data controls. OpenAI's Data Controls FAQ says that with it off, new conversations won't be used to train OpenAI models — but your old chats stay in your history either way.

Temporary Chat is the stricter option: these chats don't appear in history, don't create or update memories, and are not used to train models. Even so, OpenAI keeps them for up to 30 days for safety purposes.

Here's the part most guides skip: if you click the thumbs-up or thumbs-down button on a response after opting out, the entire associated conversation can be sent back into training. The FAQ states it plainly — feedback counts as permission. So on sensitive chats, skip the rating buttons.

Two more details worth knowing. Memory (how ChatGPT personalizes responses) and model training are separate settings — turning off one doesn't turn off the other. And if you use ChatGPT without signing in, the opt-out is saved per browser, so a new browser or device starts with training on.

Claude: your choice decides whether retention is 30 days or five years

Anthropic changed the rules in August 2025. Under the updated consumer terms, users on Claude Free, Pro, and Max are asked whether to allow their chats and coding sessions to improve Claude. If you say yes, new or resumed chats can be kept for up to five years. If you decline, the previous 30-day retention continues to apply.

Paying for Pro or Max does not buy extra privacy — the training choice is the same on every consumer plan. Only the commercial products (Claude for Work Team and Enterprise, Claude Gov, Claude for Education, and API use including via Amazon Bedrock or Google Vertex AI) are exempt from training by default.

Deleted conversations are excluded from future training runs, though anything already absorbed into a finished training run can't be pulled back out. For single sensitive questions, Claude's Incognito mode is never used for training regardless of your main setting.

Gemini: the reviewed copy outlives your delete button

For adults, Google's "Keep Activity" setting is on by default. With it on, your saved chats and uploads are used to improve Google's services, including training generative AI models, and they're auto-deleted after 18 months (you can change that to 3 or 36 months, or no auto-delete). This is spelled out in Google's Gemini Apps Privacy Hub.

The surprise is the human-review exception. Human reviewers read some chats to improve quality and safety, with the chats disconnected from your account and identifiers removed first. Chats that have been through human review are retained for up to three years — and they are not deleted when you delete your activity. Turning off Keep Activity stops future chats from being selected for review, but the copies already reviewed stay.

With Keep Activity off, or in a Temporary Chat, Google keeps your chats for 72 hours and doesn't use them for training unless you send feedback. They are still processed for safety, including by human reviewers. Google itself advises: don't enter confidential information you wouldn't want a reviewer to see.

Four traps most privacy guides skip

1. A thumbs-up can undo your opt-out. On ChatGPT, rating a response after opting out re-admits the whole conversation into training.

2. Deleting your history doesn't delete the reviewer's copy. On Gemini, chats already read by human reviewers are kept up to three years, separate from your account.

3. A paid plan doesn't automatically protect you. Claude Pro and Max have the same training choice as Free; only Work, Enterprise, Gov, Education, and API plans are exempt.

4. "Temporary" still means temporarily stored. Temporary chats live up to 30 days on ChatGPT and 72 hours on Gemini, so they pass through safety review either way.

What to check on each app: a five-minute privacy pass

ChatGPT: Settings, Data controls, turn off "Improve the model for everyone." Use Temporary Chat for one-off sensitive questions. Don't click thumbs feedback on chats you'd rather keep out of training.

Claude: Settings, Privacy, check the model-improvement choice and flip it off if you never meant to allow it. Use Incognito mode for single sensitive questions.

Gemini: turn off "Keep Activity" and consider shortening auto-delete to 3 months. Use Temporary Chat for anything personal. And review what Gemini remembers about you now and then.

The rule that never fails

Every provider has to read your prompt to answer it, so no mainstream assistant is end-to-end encrypted — the text is always decrypted on the provider's infrastructure. Settings shrink your exposure, but the only guaranteed-private prompt is the one you never typed. Whatever the settings say, strip names, ID numbers, account numbers, and passwords before pasting anything into an AI chat.

Policies change — the Claude shift happened in 2025 and settings moved on all three services in the last year — so check the actual settings page on the app you use rather than trusting an old guide, this one included.

Comments

Popular posts from this blog

AI and Jobs: Which Tasks Change First, According to Researchers

What Marketing Tasks Will AI Replace Next — and Which Still Need a Human?

How AI Screens Your Resume Before a Human Sees It — and How to Write for It