Story
When an AI product wants to use what I share to train its models, I want to be asked in plain words and find my answer again easily, so my clients' work stays where I put it.
Fatima, a freelance translator in Casablanca, trying a new AI tool on her laptop for client work.
- 01 Setting
Fatima opens a new AI tool to build a glossary for a client contract.
- 02 Trigger
The product wants to use what people share to train its AI (teach it from real examples).
- 03 Without it
A pre-ticked box at sign-up says yes for her. Her client's contract may now be used for training.
- 04 With it
A clear question, what is used, 2 equal buttons. She taps Don't allow. Her file says "Not used for training".
- 05 Outcome
Months later she finds the same choice in 1 place in Settings, 1 tap to change.
Live demo
This demo follows a script, so it plays the same every time - there is no real AI behind it. Change a setting to see the same moment play out differently.
What happens, step by step
| State | The person sees | The AI is | When | Screen reader |
|---|---|---|---|---|
| 1 | A plain question before the first chat. Nothing is chosen for her. | Waits. | first use | “Choice needed: data use” |
| 2 | What is used and what never is, in 2 short lines. | Waits. | on tap | - |
| 3 | A line confirms the choice in plain words. | Idle. | on tap | - |
| 4 | Where it matters - on a file - the choice shows again: "Not used for training". | Works. | on upload | - |
| 5 | 1 switch in Settings › Data, with what it applies to. | - | any time | - |
On each device
- phone
- The question is a full-width card or sheet. The 2 buttons sit side by side at the same size. Settings › Data is 2 taps from the menu.
- tablet
- As on desktop.
- desktop
- The question is a card in the empty chat, not a pop-up over the page. Settings › Data is linked from the account menu.
When to use it
- What people share could be used to train models, shared with others, or kept for a long time.
- People add other people's data - client files, colleagues' emails, family photos.
- Laws or your own policy need a clear yes before data is used.
- The question is about what the AI remembers about 1 person, not training. Use Memory controls instead.
- People want 1 chat to leave no trace at all. Use Temporary chat instead.
- The AI is about to act on someone's behalf - ask about that action, not about data. Use Approval gate instead.
The parts of the pattern
- APlain questionSays what will happen: "Use your chats to train our AI?" Not "Help us improve".
- BWhat is usedWhat is used and what never is, in 2 lines. The full policy sits behind a link.
- C2 equal buttonsSame size and style. Nothing chosen before the person chooses.
- DWhere to change itNames the 1 place to change it later, before people decide.
Do and Avoid
Use your chats to train our AI?
Used: messages, files. Never: temporary chats.
Don’t allowAllow
Help us improve!
Sounds goodmaybe later
📄 client-contract.pdf
Not used for training
Settings › Data
📄 client-contract.pdf
Settings › Account › Privacy › Advanced
Writing and accessibility
- Name the use: "train our AI", "share with our partners". Not "improve our services".
- List what is used and what never is, in 2 lines: "Used: your messages and files. Never used: temporary chats."
- Label the buttons with the choice: "Allow" and "Don't allow". Not "Sounds good" and "Maybe later".
- Say where to change it, and what changing it does: "Applies to new chats."
- Both buttons are real buttons with the same size, contrast and focus style. Neither has focus (the keyboard's current spot) by default.
- If the question blocks the product, it is a dialog with a clear heading and no close button that means yes.
- Switches in Settings have a visible text state ("On", "Off") and a label that names the use.
- The short version of what is used is real text, not an image, and is read before the buttons.
Seen in real products
- ClaudeWeb, desktop and mobile · Aug 2025Existing users saw a pop-up to choose whether their chats are used for training; new users choose at sign-up. The choice can be changed any time in Privacy Settings, and turning it off applies to new chats.Anthropic: Updates to Consumer Terms and Privacy Policy ↗
- ChatGPTWeb, iOS and Android · Nov 2026Settings › Data Controls holds the training choice on every platform. While it is off, new chats are not used to train models. Temporary Chats are never used for training.OpenAI Help: Data Controls FAQ ↗
- Microsoft 365 CopilotMicrosoft 365 apps · Nov 2026When people send feedback, the pane asks permission before it adds the prompt, the response or files, and lets people preview them first.Microsoft Support: Providing feedback about Microsoft Copilot ↗
Each example comes from the product’s help pages, release notes or teardowns, with the month it was checked. Products change every month - tell us when 1 is out of date.
Check yourself
Seen this pattern done better, or a product that changed? Press Suggest next to any heading, or tell us about the whole page.