In 2026, Google quietly expanded the scope of what it uses for AI training. The change, rolled out across multiple Google services, updated default settings to allow photos, audio, video, and other files uploaded to Google’s services to help train its AI systems — unless users actively changed the settings themselves.
This is an opt-out change, not an opt-in one. If you did not hear about it and have not changed anything, your personal files may already be contributing to Google’s model training.
This post explains what changed, what it covers, and exactly where to find every relevant setting.
What Changed and When
Two overlapping changes are worth understanding separately.
September 2025: Gemini training default turned on. Google updated its Consumer Terms of Service to allow conversations with Gemini — including content you share during those conversations — to be used for model training by default. This applied to Free, Pro, and Max consumer accounts. Users could opt out, but the default changed from off to on.
April–July 2026: Expanded to uploaded media. A second, broader change extended similar permissions beyond conversational data to include uploaded media: images, audio recordings, video files, and other files shared with Google services. The scope expanded beyond Gemini to include Google Search, Maps, Shopping, Flights, Hotels, Translate, and News.
The practical effect: if you upload a photo to Google Search to reverse-image-search it, that image may be retained and used for training. If you ask Google Maps about a location while logged in, that interaction becomes training data.
Google maintains that photos stored in Google Photos are not used to train external generative AI models — Gemini features inside Photos operate under separate terms. But the broader question of what Google does with media you share across its properties is now significantly more complicated than it was two years ago.
The Opt-Out Paths: A Complete Guide
These settings are scattered across different Google products with no single “privacy dashboard” that covers all of them. You need to check each one separately.
1. Gemini Activity (Conversations + Shared Files)
This controls whether your Gemini conversations — including any images, documents, or files you share during those conversations — are retained and used for training.
On the web:
- Go to gemini.google.com
- Click the gear icon or your profile → Activity
- Click the toggle next to Gemini Apps Activity
- Select Turn off and delete activity
Turning this off deletes existing saved activity and stops future conversations from being retained for training.
On mobile:
- Open the Gemini app
- Tap your profile icon → Gemini Apps Activity
- Toggle off
If you use Gemini inside Google Workspace (Gmail, Docs, Slides), those interactions are governed by your Workspace account’s terms — typically not used for training, but verify with your administrator.
2. Google Photos (Gemini Features)
Google Photos has its own settings for AI features powered by Gemini, separate from your broader Gemini Activity toggle.
On iOS or Android:
- Open Google Photos
- Tap your profile picture (top right)
- Go to Photos settings → Preferences → Gemini features in Photos
- Review and disable features you do not want active
The Gemini features inside Photos — contextual search, photo descriptions, suggested edits — send photo data to Google’s servers for processing when active. Disabling them keeps your photo library local to the app.
3. Web & App Activity (Includes Media Uploads to Search)
This is the broadest toggle. Web & App Activity governs what Google saves when you use Search, Maps, and other Google products while logged in — including any media you upload during those sessions.
On the web:
- Go to myaccount.google.com
- Click Data & privacy
- Under “History settings,” click Web & App Activity
- Toggle off, or click Turn off
- Choose whether to also delete existing activity
There is also a sub-setting within Web & App Activity: Include audio recordings. If you use voice search, spoken queries, and interactions with Google Assistant through Search are captured here. Unchecking this box stops those recordings from being retained.
4. Location History and Maps Activity
If you use Google Maps while logged in, your location data is linked to your account and can inform AI models used for mapping and local recommendations.
- Google Account → Data & privacy → Location History
- Toggle off, or manage on a per-device basis
Turning this off stops future location data from being saved. It does not delete past data — use the Manage all activity link to delete historical records.
5. YouTube Watch and Search History
If you create or upload videos to YouTube, Google’s AI systems learn from viewing and engagement patterns. The watch and search history associated with your account contributes to recommendation models.
- Google Account → Data & privacy → YouTube History
- Turn off YouTube Watch History and YouTube Search History
What About Data Already Collected?
Opt-outs control future collection. They do not automatically remove what Google has already retained.
To delete existing data:
- Go to myactivity.google.com
- Use the Delete activity by option to select date ranges
- For specific product data, filter by product before deleting
Google states that deleted data will not be used for future model training. However, if data was used for training before deletion, that training is not reversed — the model already learned from it.
This is a general limitation of opt-out systems: they stop accumulation, but they do not unwind what has already been incorporated into model weights.
Why the Default Matters
Defaults determine the outcome for most users. The majority of people do not read Terms of Service updates. They do not check settings panels when a product announces a new capability. They use the product as it came configured.
Google’s decision to make AI training opt-out rather than opt-in means that the population contributing to its model training is “everyone who didn’t change the setting” rather than “everyone who chose to participate.” At Google’s scale, that is an enormous difference.
This is not a novel complaint. Regulators in the EU have flagged similar opt-out defaults as potentially inconsistent with GDPR’s requirement for a genuine, unambiguous consent signal. Several companies have been fined or required to modify defaults as a result. The trajectory of enforcement suggests this debate is not resolved.
For now, the practical reality is that the defaults favour training, and the burden of changing them is on you.
Thinking About Where You Put Personal Files
The opt-out process above assumes you want to continue using Google services and prefer to do so with fewer data-sharing defaults active. That is a reasonable choice.
A different approach is to consider which files belong on Google’s infrastructure in the first place.
Documents you work on publicly, searches you run for general information, maps you use to navigate unfamiliar places: these are genuinely well-suited to a cloud service that benefits from broad training data.
Personal photos from your daughter’s birthday party, voice recordings of an elderly relative telling family stories, scanned copies of identity documents: these are different. Their value is personal. Their sensitivity is high. The benefits of using them to train a search engine’s AI models accrue to Google, not to you.
The question worth asking is not only “how do I opt out?” but “where should this file actually live?”
Services designed specifically around personal memory storage — rather than around advertising, search, or AI development — operate under different incentive structures. They do not need your photos to improve a search product. They have no revenue model that depends on your data reaching a training pipeline.
Knowing the difference between those two categories of service is, ultimately, more durable protection than navigating a settings panel that may look different in six months.