// AI Advantage
The 60-Second Toggle
Four settings, one per assistant, that decide whether your chats train somebody else's model. Copy the card, flip yours, send it to your team.
Quick test. Do you know, right now, whether the chatbot you used this morning is allowed to train on what you typed? I didn't, for one of mine, until I went and looked — and I write about this stuff for a living. It's not a policy question and you don't need a meeting. It's a toggle, it takes about a minute per tool, and once it's off you're the kind of person who owns their inputs instead of donating them.
Here's the card. Paste it into your team channel as-is.
AI TRAINING TOGGLES — check yours, takes 1 minute each
ChatGPT Profile > Settings > Data Controls >
"Improve the model for everyone" > OFF
(your chat history stays put)
Claude Settings > Privacy > "Help improve Claude" > OFF
Gemini myactivity.google.com/product/gemini >
Keep Activity > OFF
(trade-off: Gemini stops saving your chats too)
Copilot Profile > Privacy >
"Training on conversation activity" > OFF
(personal accounts only; work accounts aren't used
for training)
Done? Reply with a checkmark.
A few things worth knowing before you flip them.
ChatGPT is the painless one. OpenAI's data controls page says your conversations stay in your history and just stop being used for training, and the setting follows your account across devices.
Claude asks you to choose, and the choice carries more than training. Anthropic's privacy settings doc walks through the toggle, and its data usage page says consumer accounts that allow training get a five-year retention period versus 30 days for accounts that don't. That's a bigger difference than most people realize they signed up for.
Gemini is the one with a real cost. Per Google's privacy hub, turning off activity stops your chats from being reviewed or used to improve Google's models — and also stops saving them, so you lose your history. Google still holds conversations for up to 72 hours either way, and anything a human reviewer already looked at can be kept for up to three years. Decide which you want more.
Copilot only needs this on personal accounts. Microsoft's privacy controls page has the path, and its privacy FAQ says Microsoft 365 Copilot conversations aren't used to train its models.
What changes: You stop guessing what happens to the things you type, and you stop being the person who pasted a customer contract into a tool that was set to learn from it.
Where else this works: Same one-minute check on your meeting notetaker, your design tool, your CRM's AI add-on, and whatever your kids use for homework — anywhere there's a text box, there's a setting.
Why this week: Salesforce launched its first in-house model at Dreamforce, and the line it led with wasn't speed or smarts. It was that no customer data crosses its trust boundary at inference. When a company that size makes data control the headline, that's the market telling you what it values. You can't post-train a model this afternoon. You can flip four toggles. Do it now, then hit reply and tell me which one surprised you.
