privacy

Google Gemini Trains on Your Chats: What to Know

Google's Gemini app trains on your personal conversations by default — even on paid plans. Here's what their policy says and your opt-out trade-off.

If you use Google Gemini on a personal account — free or paid — your conversations may be used to train Google’s AI models. This is not a rumor. It is disclosed in Google’s Gemini Apps Privacy Hub, which was last updated on October 23 of this year. What makes it unusual is that paying for Gemini Advanced does not change this default.

The issue surfaced in a Hacker News thread in early November and quickly gathered significant attention. The core complaint is straightforward: unlike some competitors, Google offers no way to keep your chat history saved while simultaneously opting out of training. The opt-out and the deletion of chat history are coupled together.


What Google’s Privacy Hub Actually Says

Google’s privacy documentation for personal Gemini accounts states that saved conversations may be reviewed by human reviewers and used to improve Google’s AI products. This includes Gemini Advanced subscribers.

The relevant setting is called “Gemini Apps Activity.” When it is on, your chats are saved and may be used for training. When it is off, chats are deleted after 72 hours and are not used for training.

There is no third option: save chats and opt out of training.

A commenter in the Hacker News thread described the situation accurately: “With ChatGPT (even on Free) and Perplexity, I can keep my full chat history saved while separately opting out of having my conversations used for model training. With Google, you can’t.”


The Paid Plan Catch

The absence of an opt-out for training data on paid plans is the detail that frustrated users most. The common expectation — grounded in experience with other subscription services — is that paying for a product reduces how much of your data is used to subsidize that product.

Google’s arrangement is different. Gemini Advanced uses the same privacy hub settings as the free tier for personal accounts. Subscription revenue and data revenue are, in this model, not in conflict.

This is not unique to Google. Subscription products that also collect training data are a common pattern in the AI industry. What distinguishes the Gemini situation is the lack of a granular opt-out that other platforms have implemented.


What Gemini Can Access About You

The scope of what Gemini can process extends beyond the conversations themselves. Google has connected Gemini to its broader ecosystem of personal data.

If you have enabled Gemini’s Google extensions, the assistant can access and process content from Gmail, Google Calendar, Google Drive, and Google Docs. A query asking Gemini to summarize your emails or find files matching a description gives it access to some of the most sensitive personal and professional content in your Google account.

The same privacy hub that governs chat activity governs these integrations. Turning off Gemini Apps Activity limits how that activity is retained and trained on, but it also disables the history that makes conversational continuity possible.


How This Compares to Google Workspace

The personal account policy differs from what applies to Google Workspace — the paid suite used by businesses and schools.

Google states explicitly that “when using Gemini in Workspace, we do not use that personal content to train our foundational generative AI models.” Enterprise and education accounts operate under separate terms that limit Google’s use of content for model training.

The gap between consumer and business protections is significant. The same AI features that employees using company accounts can rely on with stricter data handling are available to consumers under meaningfully different terms.

If you use Google Workspace through your employer, your work interactions with Gemini are governed by the stricter Workspace policy. Personal interactions on a personal account are not.


What Human Review Means in Practice

The disclosure that conversations may be reviewed by human reviewers is worth taking seriously.

Human review in AI development typically serves a training data quality function: reviewers read selected conversations and annotate them, correct model outputs, or flag responses that need improvement. These are real people reading conversations users believed were private exchanges with an AI.

The selection process for human review is generally automated and pseudonymized — conversations are stripped of obvious identifiers before review — but the conversations themselves are read.

If you have used Gemini to draft sensitive communications, process personal information, or discuss private matters, human review means that content has the potential to be read by a Google contractor. This is not unique to Google — human review is standard practice in AI development — but it is worth knowing when evaluating what to share with any AI assistant.


What To Do

Review and change your Gemini Apps Activity setting. Open myactivity.google.com or the Gemini app settings and find the Gemini Apps Activity toggle. Decide whether the trade-off — training data for persistent history — is acceptable for your use.

Delete existing Gemini activity. Even if you turn off future retention, activity from past sessions may be in Google’s systems for a period after the setting change. You can delete existing activity from your Google Account activity page under “Gemini Apps.”

Be deliberate about what you share. Regardless of your setting choices, applying the same judgment to AI assistants that you would apply to any cloud tool is sound practice. Sensitive documents, personal health information, financial details, and private communications carry risk when processed by any third-party system.

Use Workspace if available. If your employer provides Google Workspace, using Gemini through that account gives you the stricter enterprise training policy for work-related queries.

Consider alternatives with granular controls. Some AI tools — both cloud-based and local — offer the ability to persist history while opting out of training. If this control matters to you, it is worth factoring into tool selection.


The Broader Pattern

The Gemini situation reflects a structural tension in how large technology companies build AI assistants. The models improve by processing user interactions. Users want privacy protections. Companies are incentivized to couple training data collection with the product features users want most — persistent memory, personalized responses, continuity.

Separating training data collection from product functionality, as some competitors have done, is technically straightforward. It requires a business choice to offer the separation, which carries a cost in training data quality.

Whether that separation becomes standard practice will likely depend on regulatory pressure. The EU AI Act and various national privacy laws are beginning to address the terms under which personal conversations can be used for model training. The gap between enterprise and consumer protections in Google’s own product suggests the stricter model is achievable — the question is whether it becomes the default for everyone.

For now, the setting exists. It costs you your chat history. That is the trade-off Google has decided to offer.

daftei stores personal files and memories without using them to train third-party AI models. The privacy policy is explicit on this point: your content is not used for AI training, not sold to third parties, and not used to target advertising. Files are encrypted with TLS 1.3 in transit and AES-256 at rest. The service is GDPR and CCPA compliant.

Your memories deserve better than an ad platform.

Try daftei free →
← All posts