Don't Turn Off Gmail's AI Until You Watch This

What happened

If you use Gmail in India, Google’s AI features came switched on by default. A viral post on X, from Justin Brave, says Google’s AI is going through your bank statements, tax documents and medical letters, and it gives you five steps to stop it. The same steps have been doing the rounds since last November.

I’m an auditor. My job is separating what was claimed from what was tested. So I tested every line of that post against Google’s own help pages and the court timeline. Some of it is right. Some of it is almost a year old. And one step will make you feel safe when you’re not.

Here is how it started. In November 2025, a security blog said Gmail trains AI on your mail. Google answered: no settings changed, no Gemini training. The blog walked it back. On November 11, Thele v. Google was filed in the US. Now the story is back, labelled breaking. So who’s right? Strangely, both and neither.

Where the post goes wrong

The default is real. The post says these features were enabled by default, and on that it’s right. Google’s own words: they’re off by default in four places, the EU, the UK, Switzerland and Japan (Google’s page says the European Economic Area). India is not on that list. So if you opened Gmail here, they were on. You didn’t choose that. Today’s audit word is opt-out consent: your silence counts as a yes.

The lawsuit is real, but unproven. The post says the rollout is now facing a class action. In July this year the judge dismissed it. Not because Google was cleared, but because the people suing couldn’t show Gemini had actually touched their data. They refiled in August, and Google asked to dismiss it again. The next hearing is in January 2027, a date I could only find in one secondary source. There’s no verdict, no settlement, and it only covers US users.

Step two misreads the switches. Step one, Settings, See all settings, Smart features, turn it off, is right. But that one box also takes away your Primary, Social and Promotions tabs, Smart Compose and Smart Reply. Step two calls the next setting the hidden switch and says, quote, “It only works when both are off.” That’s not how these work. Under Google Workspace smart features there are two separate switches. The first covers Google’s Workspace apps, which is where Gemini inside Gmail lives. The second covers other Google products using your mail, like Wallet, or Maps picking up a booking. Each one works on its own. If all you want is Gemini out of Gmail, turn off the first Workspace switch and keep your inbox tabs.

Step four doesn’t delete what it promises. The post says turn off Gemini Apps Activity, then delete everything, all time, and promises that removes your past history. That setting is now called Keep Activity, and if you’re over 18 it’s on by default. Google says trained reviewers, including people at its service providers, read some chats.

Step five is oversold. Proton Mail’s end-to-end encryption works between Proton users, or when you use PGP or a password-protected email. Your bank does neither. Its statement leaves the bank’s server as normal email. Proton then locks it in storage so even Proton can’t read it, and that part is real. But the bank still keeps its copy, and subject lines like “Your statement for October” aren’t end-to-end encrypted. Moving your bank, your insurer and your doctor to a new address is a week of admin, not a five-minute fix.

The evidence an auditor should ask for

For step four, the evidence is Google’s own privacy page. Chats that reviewers already saw are not deleted when you delete your activity. They’re kept for up to three years. And even with Keep Activity off, your chats stay with Google for 72 hours. So the order matters: turn Keep Activity off first, then delete. And don’t type anything into a chatbot you wouldn’t hand to a stranger.

Gmail isn’t the only account that switched something on for you. I asked three questions of each: is it on, what does it use, and can you take it back?

1. LinkedIn uses your profile and posts to train generative AI by default, Indian accounts included. Turn off Data for Generative AI Improvement. It only stops data from now on.
2. X: a box letting your posts and your Grok chats train Grok has been ticked by default since 2024. Untick it under Privacy and safety. It only covers future training.
3. Gemini on Android: since July 2025 it can work with your phone, messages and messaging apps, even with its activity setting off. In the Gemini app, go to Apps and switch off the ones you don’t use.
4. Meta AI: since December 16th, 2025, what you say to it can shape the ads and posts you see. Outside the EU, the UK and South Korea there’s no off switch. Meta says it won’t use topics like health or religion for ads. We can’t test that, so don’t tell Meta AI anything personal.
5. Sign in with Google: each app you signed into may have asked to read your mail or your Drive. I found 18 connected accounts, the oldest from 2017.

The test to run tonight

Do this in this order:

1. Gmail: turn off the first Workspace smart features switch.
2. Gemini: Keep Activity off, then delete.
3. LinkedIn, then X.
4. Gemini’s app connections.
5. Your Google Account, Security, connections to third-party apps: remove anything you don’t recognise.

The verdict on the viral post: the default is real, and India is in it. The lawsuit is real, but unproven and a year old. Step one works. Step two works, but not for the reason it gives. Step three works. Step four doesn’t delete what it says it does, and step five is oversold.

A default is a decision someone else made for you. Don’t audit it only when a post scares you. Put a reminder in your calendar every six months and run this list again.