Is Gmail training AI on your email? What actually changed

Over the past few weeks a claim went around that was hard to miss: Google had quietly opted everyone in to letting its AI read and train on your private email, and you needed to go and switch it off immediately. It came with screenshots, an urgent tone, and a lot of shares.

Enough people asked us about it that it is worth writing down properly.

The short version

• Did Google change your settings? No.
• Is Gemini trained on your email? Google says no.
• Is Google software reading your email? YES! Gmail's smart features have scanned your mail for years, to power reply suggestions and package tracking.
• What changed in July? Two words in a policy document.

Everything below is the detail behind those four answers.

What actually happened

In July, Google updated its privacy policy. In the section describing how it uses your information, the phrase "language models" was replaced with "AI models," and the policy spelled out that publicly available information, meaning material already public on the web rather than anything sitting in your inbox, can be used to build features and products.

That is the change. A wording update to a policy document. It is the kind of edit that would normally pass unnoticed, and this time somebody noticed, screenshotted it, and the interpretation ran a long way ahead of the text.

Your settings were not changed. Google has stated it did not alter anyone's settings. Whatever you had switched on before, you still have switched on. Nothing was flipped while you slept.

Google says Gemini is not trained on your inbox. A Google spokesperson called it misleading to say the company trains its Gemini model on users' emails. Fact-checkers at Snopes looked into the claim in August and reached the same conclusion.

And here is the part the panic got right. Your mail is being read by software, and it has been for years. Gmail's "smart features" scan the contents of your messages. That is how you get reply suggestions, package tracking that appears on its own, calendar events pulled out of confirmation emails, and filtered categories. None of that works without something reading the mail.

So the honest summary is: no new grab happened, and an old one is still running.

The two settings worth checking

Neither of these is dramatic, and you may decide you want the features. That is a fair trade to make on purpose. The problem is making it by accident.

1. Gmail smart features

1. Open Gmail, go to Settings, then See all settings.
2. On the General tab, look for "Smart features and personalization".
3. Switch it off and save.

What you give up: reply suggestions, package tracking that appears on its own, calendar events pulled out of confirmation emails, and the tabbed categories that sort your inbox.

What you get back: Gmail stops scanning the contents of your messages to power those features.

That is the whole trade. Convenience on one side, your mail not being read on the other.

2. Keep Activity, previously called Gemini Apps Activity

1. Go to myactivity.google.com.
2. Open the Gemini Apps section.
3. Turn off "Keep Activity".

What you give up: your chat history. Past conversations stop being saved, so you cannot go back and find one, and Gemini stops tailoring answers to what you asked before.

What you get back: your conversations are no longer kept and reviewed by people to improve the service.

Worth knowing: Google says that even with this off, chats are still held for up to 72 hours so the service can run and handle feedback. Off means not kept long term. It does not mean never stored.

Both take about a minute. Neither requires deleting anything or leaving Google.

If you are not on Gmail

You are not exempt. Yahoo, AOL and Outlook all run their own AI-powered features across your inbox, and they each have their own settings buried in their own menus. The specifics differ. The shape of the arrangement does not.

The part that actually matters

This particular scare was wrong. What it revealed is not.

Look at what had to happen for any of this to be a story. A company changed two words in a policy document. Millions of people had no way to know whether that mattered, no practical way to check, and no ability to say no without abandoning the account their entire digital life is attached to. The uncertainty is the point. When your data lives in someone else's building, the terms can be rewritten on their schedule, and your options are to accept it or to leave.

Most of the time the change is harmless. Sometimes it is not. Either way, you find out afterwards.

That asymmetry is the reason we build what we build. Not because Google is uniquely villainous, but because "trust us, and check back after every policy update" is a strange arrangement to have with a decade of your family's photos, messages and documents.

The panic was wrong this time. The uneasiness is not.