How to Turn Off Gmail's AI: Every Switch, Including the Two Most People Miss

Last updated: June 2026

If a post about Gmail's AI reading your bank statements and medical letters landed in your feed this week, here is the short version: the settings are real, the lawsuit is real, and the scariest claim is not. Google states plainly that it does not use Gmail content to train its Gemini AI model. What the company's smart features do is read and process your email content to power things like inbox tabs and package tracking, and a federal class action alleges Google switched that processing on by default for US users without consent.

Turning it off takes more than one click. The controls sit in at least four separate places: two settings on the Gmail desktop site, separate per-account settings in the mobile app, and a Gemini activity page most people have never opened. This guide walks through every switch, explains what each one controls and what you give up, covers the parts no setting can turn off, and ends with the one move that takes your most sensitive email off scanning infrastructure entirely.

What Actually Happened: The Default Flip and the Lawsuit

On November 11, 2025, an Illinois Gmail user filed a proposed class action against Google. The case, Thele v. Google LLC, No. 5:25-cv-09704 in the Northern District of California, alleges that the company switched on Gemini-powered smart features for Gmail, Chat, and Meet accounts by default, from at least October 10, 2025, without users' knowledge or consent.

According to the complaint, the change gave Gemini access to "the entire recorded history of its users' private communications," including every email and attachment in Gmail. The suit brings claims under the California Invasion of Privacy Act, the federal Stored Communications Act, and California's computer data access law. The California privacy statute matters here because it requires consent from all parties to a communication and carries statutory damages per violation. Legal analysts note that earlier Gemini integrations were opt-in, and the complaint's core theory is that a silent default flip is not consent.

Two things to keep straight. First, these are allegations, and nothing has been proven. Google has asked the court to dismiss the case, arguing the plaintiffs do not claim their own data was actually accessed or that they suffered any concrete harm. As of June 2026, that motion is pending: no ruling, no class certification, no settlement. Second, the lawsuit landed in the middle of a viral panic, and the panic's loudest claim was wrong.

What Google's Denial Says, and What It Doesn't

On November 21, 2025, Google published a three-point response to what it called misleading reports: it had not changed anyone's settings, smart features have existed in Gmail for years, and "We do not use your Gmail content to train our Gemini AI model," the company stated. Malwarebytes, the security firm whose article helped drive the panic, corrected its own reporting after reviewing Google's documentation. On the training question, the record favors Google.

Notice what the denial does not say. It does not say smart features avoid reading your email, because they cannot work without reading it. Package tracking means parsing receipts. Inbox tabs mean classifying message content. Calendar auto-add means extracting flight details from confirmations. The legal question in Thele is not whether Gemini trains on your mail. It is whether running AI analysis across private messages, enabled by default, with the off switches split across multiple screens, is something users meaningfully agreed to.

One Inbox, Two Privacy Standards

In the European Union, the United Kingdom, Switzerland, and Japan, Google ships these smart features off until the user opts in. In the United States, reporters and users checking their accounts in November 2025 found them switched on, and Google's own help page attributes the regional difference to those regions' privacy laws. Same product, same company, same AI. The only variable is whether the law lets Google choose your default. Whoever controls the infrastructure controls the experience, and a default setting is infrastructure.

Every Switch at a Glance

Four locations, in the order you should work through them. The detailed steps for each follow below.

Paths verified June 2026. Google periodically renames and relocates these settings; if a menu does not match, search the settings page for "smart features."

Switch 1: Smart Features in Gmail, Chat, and Meet

This is the switch Google surfaces in its own messaging, and it is where most people stop.

1. Open Gmail in a desktop browser, click the gear icon in the top right, then See all settings.
2. On the General tab, scroll to Smart features and personalization.
3. Uncheck Turn on smart features in Gmail, Chat, and Meet.
4. Scroll to the bottom and click Save Changes. Gmail reloads.

What you give up, honestly: the Primary, Social, and Promotions tabs collapse into a single inbox, Smart Compose and Smart Reply disappear, and automatic package and order tracking stops. Manual filters, labels, and search operators keep working. Your inbox gets dumber, not broken.

Switch 2: The Workspace Setting Most People Miss

Turning off the first setting does not touch this one. Google's own documentation treats the two as independent controls, which is why so many people who think they opted out have not.

1. On the same settings page, scroll to Google Workspace smart features.
2. Click Manage Workspace smart feature settings.
3. Turn off Smart features in Google Workspace.
4. Turn off Smart features in other Google products.
5. Save.

Per Google's documentation, the first checkbox governs whether your Workspace content, meaning Gmail, Drive, and Calendar data, feeds features across Workspace apps: flight confirmations landing on your calendar, file suggestions in Drive, smarter search results. The second governs whether that same content personalizes products outside Workspace: restaurant reservations surfacing in Maps, tickets and loyalty cards in Wallet, and personalization in Gemini and Search. Leave either box checked and a pipeline stays open.

Switch 3: Your Phone Has Its Own Switches

Mobile settings do not reliably mirror the desktop, so check the Gmail app on every device you use, for every account signed in on it.

1. Open the Gmail app and tap the menu or your profile picture, then Settings.
2. Select an account. Repeat these steps for each one.
3. Find Smart features and personalization, which sits under a Data privacy heading on some versions, and turn it off. Confirm when prompted.
4. On the same screen, open the Google Workspace smart features controls and turn those off as well.

Labels differ slightly between Android, iOS, and app versions; if a menu does not match, search the settings screen for "smart features." And take the per-account part seriously. A locked-down primary address means little if an old secondary account on the same phone is still running defaults.

Switch 4: Gemini Apps Activity, Off and Deleted

If you have ever used the Gemini chatbot, even once, your conversations are saved to your Google account by default for 18 months, and a sample of them can be read by human reviewers.

1. Go to myactivity.google.com/product/gemini and sign in.
2. Turn off Gemini Apps Activity. Google labels this Keep Activity on some accounts. If offered the option to turn off and delete activity, take it.
3. Select Delete, then All time, to clear past history.
4. If you would rather keep history on, set Auto-delete to the shortest window, three months.

What No Setting Turns Off

This is the part the viral threads skip, and it comes straight from Google's Gemini Apps Privacy Hub.

The 72-hour floor. Even with activity turned off, Gemini conversations are stored for up to 72 hours so Google can run the service and process feedback. There is no zero-retention configuration on the consumer product.

The three-year shadow. Conversations selected for human review are retained for up to three years, stored separately from your account. Deleting your activity does not delete them. Google's own guidance tells users not to enter confidential information into Gemini conversations, and that advice is worth taking literally.

Security scanning continues. Spam, phishing, and malware filtering run regardless of smart-feature settings, on a separate pipeline. That is a good thing; those scans protect you. The switches above govern feature processing and personalization, not security.

There is also a structural point hiding in all of this. Every control on this page is policy, not architecture. The defaults flipped once for US accounts while staying off in Europe, and they can flip again with the next terms update. A privacy posture that depends on rechecking four menus after every product announcement is not a posture. It is a chore.

The Durable Fix: Move Sensitive Email Off Scanning Infrastructure

A provider cannot scan what it cannot read. Proton Mail stores messages with zero-access encryption: content is encrypted so that Proton itself cannot read your stored mail, and messages between Proton users are end-to-end encrypted. The privacy is architectural rather than promised. There is no setting for the provider to flip later, because the capability to read your inbox does not exist on their side. The service is Swiss-based, independently audited, and runs no advertising business that would benefit from your message content.

The practical pattern is a split inbox. Keep Gmail for newsletters, receipts, and sign-ups, the mail you would not mind a model summarizing. Move medical, legal, financial, and tax correspondence to Proton, and update those senders. Proton's import tool can carry over your existing Gmail history. The honest tradeoffs: searching encrypted message bodies is more limited than Gmail's server-side search, and the migration takes a deliberate afternoon rather than a click. The free tier is enough to test the entire workflow before committing to anything.

Get Proton Mail: Free Tier Available

And if what you actually like about Gmail's AI is the features, the summaries and drafting and quick answers, you do not need a cloud scanner attached to your inbox to get them. Google's own Gemma 4 models are open-weight and run on hardware as small as a Raspberry Pi, and a mid-tier machine from our local AI mini PC guide runs the larger variants at conversational speed. A local model reads exactly the documents you hand it and nothing else, and it pairs well with network-wide tracker blocking through Pi-hole. That is the local-first version of this story: AI as a tool you point, not a service that watches.

Frequently Asked Questions

Does Gmail use your emails to train Gemini?

No. Google's November 2025 statement says Gmail content is not used to train the Gemini model, and no credible reporting has contradicted that. Smart features do process your email content to operate, classifying, extracting, and suggesting, which is what the consent lawsuit is about. Training and processing are different questions, and the viral posts collapsed them.

What is Thele v. Google, and has anything been decided?

It is a proposed class action filed November 11, 2025 in the Northern District of California, alleging that Google enabled Gemini smart features by default from at least October 10, 2025 in violation of California and federal privacy law. Google has moved to dismiss, and as of June 2026 there is no ruling, no class certification, and no settlement. There is also no claim form: any site or message asking you to pay to join a Gmail AI lawsuit is a scam.

Why are smart features opt-in in Europe but on by default in the US?

Google's help documentation attributes the regional split to privacy law. The European Union, United Kingdom, Switzerland, and Japan require opt-in consent for this kind of data processing, so Google ships the features off there. United States law does not require it, and US accounts have been found with the features on. Where regulators force the choice onto the user, the default protects the user; where they do not, it has not.

Will turning off smart features break my inbox?

No, but it will simplify it. You lose the Primary, Social, and Promotions tabs, Smart Compose and Smart Reply, automatic package tracking, and calendar auto-add from emails. Mail delivery, manual filters, labels, and search all keep working exactly as before.

Does turning off smart features stop spam filtering?

No. Spam, phishing, and malware scanning belong to Gmail's security pipeline, which runs regardless of smart-feature settings. Turning off smart features changes what Google does with your content for features and personalization, not what it does to protect your account.

Is Proton Mail actually more private than Gmail?

For stored content, yes, and by design rather than by policy. Proton's zero-access encryption means the provider cannot read message bodies sitting in your mailbox, while Gmail's privacy model depends on what Google's current policies permit. The tradeoffs are real: more limited search of encrypted content and a migration effort. For medical, legal, and financial mail, that exchange favors encryption.