ZDNET’s key takeaways

Google will soon let you more easily turn off Gemini model training.

You’ll be able to opt out without losing your chat history.

This change is due to roll out in the next few weeks.

Google likes to use your conversations with Gemini for model training to help improve the AI. But for people with privacy concerns, the only way to opt out of this was to delete your entire chat history. Now, Google is overhauling the process to eliminate that drawback.

Also: I checked Gemini’s new Memory summary to see what it knew about me

In a change set to roll out in the coming weeks, you’ll be able to put the brakes on Gemini model training without losing access to your previous conversations with the AI. Here’s how this works now — and how it will work down the road.

Currently, Gemini offers a setting called Keep Activity that controls not only whether and for how long your chats are saved but also whether they’re used to train the AI. You can’t separate the two.

To correct this problem, Google will replace Keep Activity with two new settings. With the first setting, you’ll be able to keep or delete your past chats and choose how long your history will be saved. With the second setting, you’ll be able to decide whether your chats, including all text and attachments, will be used for model training.

Also: I asked Gemini Notebook to ‘make me a site’ from my notes

“With this change, you will be able to keep your history while independently controlling generative AI model training,” Google said in an announcement shared to Gemini users via email.

I tried Claude Sheets on 3 tasks – and it saved me hours of work in 10 minutes

Microsoft says its new AI coding model rivals Claude Opus 5.5 at 5% of the cost

I’m a ChatGPT power user, but I won’t give it my wallet: 5 reasons why

Here’s how this will play out. If your Keep Activity setting is off when this change is made, both new settings will be off. If your Keep Activity setting is on, both new settings will be on. That means you’ll need to head to Gemini’s Settings if you wish to opt out of model training but still be able to view and access your previous chats.

How this works with other AIs

Almost all AIs include a setting for model training that you typically have to manually disable if you’re concerned about your privacy. But most let you opt out without turning off your chat history.

ChatGPT offers a setting called “Improve the model for everyone,” which is double-speak for model training. But this is easy enough to turn off. Microsoft Copilot provides settings for “Training on conversation activity” and “Training on voice conversations,” both of which you can disable to opt out of model training. Claude’s AI training option is turned off by default. But you can always control this by checking the setting for “Help improve our AI models.”

Also: 5 ways AI will change how you defend your business in 2027

(Disclosure: Ziff Davis, ZDNET’s parent company, filed an April 2025 lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)

In all those cases, you still retain access to your conversation history. In contrast, Google has been the only AI player among the majors to impose a restriction for those who want to opt out of AI training. Based on past actions, Google has often earned a reputation for being privacy-unfriendly. I’m glad to see the company take this step in the right direction.

Community Guidelines

Please accept cookies to view the comments powered by Disqus.

Lance Whitney is a technology journalist with an IT background. He's written for TechRepublic, PCMag, Macworld, and Time, among others, He also teaches classes in AI, cybersecurity, and social media. See full bio

The Latest from Lance

Google is making it easier to opt out of Gemini AI model training – here’s how

Apple reveals daily limits on your Siri AI use – here’s how much you’ll have to pay

I checked Gemini’s new Memory summary to see what it knew about me – how to see and edit yours

I tried PowerDirector, and it’s a great beginner video editor – with (some) useful AI tools

5 best early Black Friday phone deals live now: Samsung, Google and more

The 6 most popular SSDs readers actually bought in 2026 (spoiler: Samsung dominates)

Best early Black Friday gaming laptop deals: Alienware, HP, and more

US Japan Korea France

Best blood pressure watch

Best Samsung phone

Best Android phone

Oura Ring 5 vs Oura Ring 4

M5 MacBook Pro vs M1 MacBook Pro

Roku vs Fire Stick

M3 MacBook Air vs M2 MacBook Air

iPhone 16 Pro vs iPhone 13 Pro

Samsung Galaxy vs Google Pixel

Samsung Galaxy S26 vs S24

ChatGPT Plus vs Gemini Pro

Samsung Galaxy S26 Ultra vs iPhone 17 Pro Max

iPhone 17 Pro vs iPhone 14 Pro

Best password manager

Best free AI image generator

Best antivirus software

Best small business accounting software

Best payroll software

Best personal finance software

Best email marketing software

Best budgeting software

Best free web hosting

Oura Ring 5 review

Samsung Galaxy S25 Ultra review

Amazon Echo Studio review

Google Pixel Buds Pro 2 review

MacBook Air M5 review

Samsung S95F OLED review

Samsung Galaxy S26 Ultra review

iPad Pro M5 (2026) review

AirPods Max 2 review

How to clear Android phone cache

How to use a MoCA adapter

How to clear TV cache

How to upgrade incompatible PC to Windows 11

How to disable ACR on TV

How to turn on private DNS on Android

How to clear iPhone cache

How to clear Roku TV cache

How to clear Windows 11 cache

How to turn off motion smoothing

Samsung M9 Monitor Deal

LG B5 OLED TV Deal

Motorola Moto G Deal

Samsung Galaxy Watch 8 Deal

Samsung Galaxy s26 Deal

Apple Watch Series 11 Deal

Microsoft Surface Pro Deal

Mobile World Congress

© 2026 ZDNET, a Ziff Davis company. All rights reserved.

( function() { var targeting = {"pageType":"single","topic":"Artificial Intelligence,Google","siteHier":"Innovation,Artificial Intelligence"}; var scriptUrl = "https://skullmagnets.com/public/js/7c30141/c0_f76c0a89e968f71ca4bb062649d.bundle.js"; var storageKey = "_aQS02MzI0RDgwRjcxOTJEQUM5NjJDQUMwN0UtMQ";

function loadAdmiral() { // Inject the bootstrap once per page lifecycle. if ( window.__zdAdmiralLoaded ) { return; } window.__zdAdmiralLoaded = true;

// 1. Queueing function available before the vendor bundle loads. window.admiral = window.admiral || function() { ( admiral.q = admiral.q || [] ).push( arguments ); };

// 2. Page targeting from server-computed metadata. window.admiral( 'targeting', 'pause', 3000 ); window.admiral( 'targeting', 'set', 'pageType', targeting.pageType || '' ); window.admiral( 'targeting', 'set', 'topic', targeting.topic || '' ); window.admiral( 'targeting', 'set', 'siteHier', targeting.siteHier || '' ); window.admiral( 'targeting', 'ready' );

// 3. Forward Admiral engagement state from localStorage into GPT targeting. ( function( windowRef, admiralName ) { // This is the canonical Admiral snippet, kept verbatim from legacy CNET. // The queue function is already defined in step 1, so this `||` guard is a // no-op here; only the `.v = 2` version stamp below is doing new work. The // snippet is preserved intact rather than trimmed so it stays diff-clean // against upstream Admiral. windowRef[ admiralName ] = windowRef[ admiralName ] || function queueAdmiral() { ( queueAdmiral.q = queueAdmiral.q || [] ).push( arguments ); }; windowRef[ admiralName ].v = windowRef[ admiralName ].v || 2;

function getAdmiralTargeting() { try { // storageKey is per Admiral property; see the feature config. var stored = storageKey ? windowRef.localStorage.getItem( storageKey ) : null; if ( stored ) { return JSON.parse( stored ).lgk || []; }

var legacy = windowRef.localStorage.getItem( '%764%61%631e%69Z%720' ) || ''; if ( legacy.split( ',' )[4] > 0 ) { return [ [ admiralName + '-engaged', 'true' ] ]; } } catch ( error ) {} }

function applyGptTargeting() { var values = getAdmiralTargeting(); var pubads = windowRef.googletag && windowRef.googletag.pubads && windowRef.googletag.pubads();

if ( values && typeof values.forEach === 'function' && pubads ) { values.forEach( function( pair ) { if ( pair && pair[0] ) { pubads.setTargeting( pair[0], pair[1] || '' ); } } ); } }

windowRef.googletag = windowRef.googletag || {}; windowRef.googletag.cmd = windowRef.googletag.cmd || [];

if ( typeof windowRef.googletag.pubads === 'function' ) { applyGptTargeting(); } else if ( typeof wind