Most major platforms now use user-generated content to train or improve AI models. The default is usually opt-in — your content is included unless you take action to exclude it.
This guide covers the specific steps to opt out on the platforms where it’s possible, flags where it’s not, and explains the limits of each approach. None of these opt-outs are retroactive, and not all of them are complete — but doing them is better than not.
Google has multiple AI training pipelines, and they have different opt-out mechanisms — or none at all.
Google Search (uploaded media)
A June 2026 policy change made images, audio, and video you submit to Google Search — through Google Lens, voice search, or image search — eligible for AI training by default.
To opt out:
- Go to your Google Account at myaccount.google.com
- Select Data & Privacy
- Under “History settings,” open Search Services History
- Toggle off Save Media
- Also check Search Services Personalization and turn it off if present
Gmail (Smart Features and Personalization)
Google’s Smart Features in Gmail allow Gemini and other AI tools to analyze your email content to power suggestions, summaries, and smart replies. These features also feed into personalization systems.
To disable:
- In Gmail, go to Settings > See all settings
- Click the General tab
- Scroll to Smart features and personalization and select Turn off
- Also visit Settings > See all settings > Chat and Meet to review connected AI features
- In Google Workspace (if you use it through an employer or school), these settings may be controlled by your administrator
Google Photos
Google Photos uses your photo content for AI-powered features like face grouping and memory creation. Separate from AI training, these features process your content server-side. Google’s updated terms for consumer accounts do not provide a specific opt-out for AI training on Photos content beyond what’s controlled by the general Smart Features setting.
If you want to remove content from Google Photos that may already have been processed:
- Use Google Takeout (takeout.google.com) to export your photos before deletion
- Delete photos you don’t want in Google’s systems via the Photos interface
- Go to the Trash and empty it — deleted photos remain in Trash for 60 days before permanent deletion
Google Workspace (Docs, Slides, Sheets)
Smart Features covers Workspace AI. Additionally, if your Workspace account has Gemini features enabled by an admin, your documents may be processed for AI functionality. Check with your organization’s administrator for Workspace-specific AI training opt-outs.
Meta (Facebook and Instagram)
Meta’s opt-out process is one of the more friction-heavy ones, and the outcomes are uncertain. Meta does not offer a simple toggle for AI training in the US. The process is a formal objection that Meta reviews individually.
For users in the United States
- Go to facebook.com/help/contact/510058597920541 (Meta’s generative AI objection form)
- Fill out the form specifying your objection to the use of your content for training Meta’s generative AI models
- Submit — Meta is required under their own policy to respond, but the process is not automatic or guaranteed
Note: This specifically covers generative AI training. Meta’s broader data processing for ads, ranking, and other systems operates under different terms.
For users in the European Union
EU users have stronger rights under GDPR’s legitimate interests provisions. Meta provides an opt-out for EU users under their privacy settings:
- Go to Facebook Settings > Privacy > Meta Privacy Center
- Look for the AI at Meta section
- Select See your privacy choices and follow the prompts for generative AI data use
The EU process is more standardized because regulators have required it. Meta has faced enforcement action in the EU over AI training data practices, which has produced more accessible controls than exist for US users.
For users in the UK and other regions
The availability of controls varies by region depending on applicable privacy law. UK users following Brexit now operate under UK GDPR, which provides similar rights to the EU version. Check Meta’s Privacy Center for your region-specific options.
What Meta doesn’t let you opt out of
Meta continues to use content for ad personalization, ranking systems, and various non-generative AI features regardless of the generative AI objection. The objection process is specifically about training generative AI models (like Llama and Muse Image), not Meta’s broader AI infrastructure.
OpenAI (ChatGPT)
ChatGPT offers one of the cleaner opt-out mechanisms of the major platforms.
For ChatGPT users
- In ChatGPT, go to Settings (your profile icon > Settings)
- Click Data Controls
- Toggle off Improve the model for everyone
When this setting is off, your conversations are not used for training. This applies to logged-in accounts; ChatGPT’s training data policies for unlogged access may differ.
For sensitive conversations
Even with training disabled, your conversations are stored unless you specifically delete them or use Temporary Chat:
- Use Temporary Chat (available in the interface) for any conversation you don’t want stored at all. Temporary chats are not used for training and are deleted within 30 days.
- Delete conversation history via Settings > Data Controls > Delete All Chats
For API users
If you use ChatGPT via the API (building applications, not using the consumer product), API inputs and outputs are not used for training by default unless you’ve opted in via the API platform settings. This default is different from the consumer product — API data is excluded from training by default.
ChatGPT Enterprise and Education
These versions have explicit data-processing agreements that exclude content from training. If you’re using ChatGPT through an employer or institution, your organization’s agreement determines how your data is handled.
Microsoft (Copilot, Bing, Windows Recall)
Microsoft’s AI features span multiple products with different controls.
Bing AI (Microsoft Copilot search features)
- Go to account.microsoft.com/privacy
- Under Privacy settings, look for AI model improvement or Bing AI training
- Toggle off sharing with Microsoft for product improvement
Windows Recall
Windows Recall is a feature that takes screenshots of your PC activity on an ongoing basis and indexes them with AI for search. As of mid-2026, it is available on Copilot+ PCs and can be disabled:
- Go to Settings > Privacy & Security > Recall & Snapshots
- Toggle off Save snapshots
Recall has been controversial since its announcement for capturing sensitive content (banking sites, personal messages) in its screenshots. If you have a Copilot+ PC, verifying this setting is off is worth doing as a baseline.
Microsoft 365 Copilot (for work and personal accounts)
Microsoft 365 Copilot processes your documents, emails, and meetings within Microsoft’s compliance boundary. Microsoft’s documentation states that Copilot does not use customer data to train foundation models. If you have concerns about specific processing, review your organization’s Microsoft 365 data governance settings or contact your IT administrator.
Apple
Apple’s AI features, branded as Apple Intelligence, run predominantly on-device for most tasks. Where server-side processing is used (for requests that exceed on-device capability), Apple routes them through Private Cloud Compute, which is designed to process requests without Apple being able to access the data.
Apple does not use Siri requests or Apple Intelligence interactions to train its foundation models, per its published documentation. There is no explicit opt-out required for training because the policy states it doesn’t happen.
However, if you want to disable Siri analytics sharing:
- Go to Settings > Privacy & Security > Analytics & Improvements
- Toggle off Improve Siri & Dictation
Platforms With No Meaningful Opt-Out
Some platforms don’t provide usable opt-outs:
Reddit — licenses posts and comments to Google and OpenAI. Individual users have no mechanism to exclude their posts from this licensing.
TikTok — the privacy policy allows broad use of content for model training. There is no individual opt-out available.
LinkedIn — has had multiple controversies over AI training opt-outs. As of mid-2026, LinkedIn provides some opt-out controls in the EU but limited options elsewhere. Check Settings > Data Privacy > Data for Generative AI Improvement.
Snapchat — content you send may be used for AI feature improvement. Settings are available under Privacy Controls > Personalization but the scope of what’s covered is limited.
What Opting Out Doesn’t Do
Opting out of AI training is not the same as:
Deleting what’s already been used. If your content was processed before you opted out, those training runs are not reversed. Machine learning models are not easily “unlearned” — removing training data doesn’t necessarily remove the model’s behavior patterns derived from that data.
Controlling all processing. AI training is one specific use of your data. Platforms separately process content for ad targeting, content ranking, spam filtering, recommendations, and other systems — opting out of training doesn’t affect those.
Protecting content you’ve already shared publicly. On platforms where content is public, others can download and use it independently of the platform’s own policies. Opting out of a platform’s AI training doesn’t prevent third parties from scraping public content.
Where Private Storage Fits In
The opt-out processes described here all apply to platforms where you’ve already shared or uploaded content. The more fundamental control is keeping certain content off these platforms in the first place.
A private personal archive — photos, documents, voice notes, files you want to keep but don’t intend to share — doesn’t face these choices, because the content never reaches a platform with AI training pipelines to opt out of.
daftei stores files with AES-256 encryption at rest and TLS 1.3 in transit. It never uses your content to train third-party AI models, never sells your data, and never shows ads. The content you store there is not subject to the opt-out processes in this guide — because it’s not in the systems those processes apply to.