It’s been a busy month as far as Gemini is concerned, with Google rolling out a whole new class of Gemini models along with a rebranding of the Gemini Spark agent, which is now called Tasks. But there’s also a new, under-the-radar setting that all Gemini users should enable, right now.
The setting is one that ChatGPT and Claude users are likely well aware of: the ability to block AI models from training on the data in your chats. Opting out of sharing your chats with OpenAI and Anthropic for AI training purposes is easy with ChatGPT and Claude; you just go to Settings and then Data Controls (for ChatGPT) or Privacy (for Claude) and disable the “help improve our models” option.
But up until now, opting out of model training for Gemini users hasn’t been so simple. That’s because Google used to link the “help improve our models” toggle with your Gemini app history, which keeps track of your Gemini chats and allows you to pick up your Gemini conversations where you left off. To turn off one, you had to also turn off the other, which meant opting out of model training also meant losing saved Gemini conversations. Yep, it was weird.
Welcome to another edition of Prompt Mode, your weekly AI newsletter.
I’m your host, Ben Patterson. Each week on Prompt Mode, I’ll be serving up analysis of the AI trends that matter to everyday users like you and me. Stay tuned for practical AI tips, hands-on experiences with the latest AI tools, and—you guessed it—prompts to help you get the most out of your AI assistants.
Thanks for reading, and if you like what you see, just sign up right here.
Now all that’s changed. If you check out your Gemini History now, you should see separate controls for your Gemini History and model training, allowing you to keep your chat history enabled while opting out of sharing your conversation data for model training. Google says the changes will be rolling out in the coming weeks, so if you don’t see the separate settings yet, you should soon.
Just click here to get to your Gemini History settings, make sure the blue toggle for History is set to On, go to the “Help train generative AI models with” heading, and uncheck “Chat content including text & attachments.” While you’re at it, you may as well uncheck “Audio and Gemini Live videos & screenshots,” too.
Why do we care about our chats being used as AI training data? Because even with the massive data sets used by ChatGPT, Claude, and Gemini, it’s possible for bad actors to make AI models leak their training data using malicious prompts, and that data could include user credentials, email addresses, private records (like, say, legal contracts or medical diagnoses), and other PII (personally identifiable information) you’ve shared in an AI chat.
The biggest AI providers say they’ll do their best to strip out any PII from your AI conversations before using them as training data. Still, better safe than sorry, which is why you should opt out of model training for all your AI accounts.
Now, opting out of model training doesn’t mean your AI providers won’t retain your chats for a certain period of time, even after you’ve deleted them, and it’s notable that Google just changed its policy here. Instead of just keeping your chats for 72 hours, Google may now keep them for “around two months” for “security and safety purposes,” including to help “detect, report, and investigate complex emerging abuse scenarios.”
That’s about twice as long as the data retention policies for OpenAI and Anthropic, which both say they may keep your chats for up to 30 days. Of course, both providers also carve out exceptions for certain “security and legal” considerations, in which case chats may be kept even longer. And given the increased (and justified) scrutiny of the latest frontier models plus the impending leadership shakeup in Congress, OpenAI and Anthropic may soon increase their data retention policies, too.
So remember: While opting out of model training for your AI chats is a wise privacy move, it doesn’t make your AI conversations completely private.
More in AI this week
- Don’t want AI in your Canva workflow? Now there’s a toggle that hides Canva’s AI tools. (9to5Mac)
- The sudden downfall of ex-Nvidia CEO Jensen Huang has spawned countless AI slop memes, including one recreating that now-famous scene from You Can See Everything. (PCWorld)
- Amid all the new Gemini models, there’s yet another revamped version of OpenAI’s flagship, with GPT-6.2 Astra (along with GPT-6.2 Sol) promising a “new quality bar. You know the drill. (OpenAI)
- The Cambridge Dictionary has named its word of the year, and it is “human,” defined as “having the qualities, faults, and feelings that people have, as opposed to gods, animals, or machines.” Take that, AI. (University of Cambridge)
- From tracking down large files to listing your installed apps, ChatGPT can create all kinds of helpful PowerShell scripts for willing Windows users. (PCWorld)
- Take a gander at our review of GPTZero, one of the most popular AI detectors. Spoiler alert: We liked it. (PCWorld)
Prompt of the week: The Amazon Black Friday price-check prompt
Want to make sure that Amazon Black Friday discount is really the best deal around? Here’s an agentic AI prompt that can help track down even more savings — and no, you won’t need to put your credit card at risk. This Amazon “price check” prompt asks ChatGPT dots, Claude Duet, Meta Muse, and other AI agents to scour the web for deals from Amazon’s competitors.
Just paste the name of the product and Amazon’s price into the prompt, and your agent will perform a read-only search for steeper discounts, all without submitting any personal information or trying to buy anything.
That’s all for now!
Thanks for reading the latest issue of Prompt Mode.
Want more next week? Don’t forget to sign up to start receiving this newsletter in your inbox.


