Anyone using Gemini previously faced a trade-off: keeping chats often meant allowing Google to use them for training its AI systems. With an update from November 16, 2024, Google now separates these two points more cleanly. For Gemini data privacy, this is a noticeable step forward, but not a free pass. In the future, chats without history will remain stored for around two months instead of the previous 72 hours. We show you what specifically is changing, which data Gemini collects, how to configure the settings sensibly, and what companies in the EU should pay attention to.
Preliminary note: Handle your data with care, as it can never be 100% guaranteed which information might flow to external services. This article describes the status as of November 20, 2024, and does not replace a case-by-case assessment.
The Most Important Points in 30 Seconds
New: History and AI training can be controlled separately in Gemini. Google has been rolling out the change gradually since November 16, 2024.
Important: Google now stores chats without history and temporary chats for around two months instead of the previous 72 hours.
Please note: Chats reviewed by humans can remain stored for up to three years, even if you delete your history.
For companies, the account type matters: A Pro subscription does not create separate data privacy rules. Binding commitments primarily exist with Workspace accounts.
First step: Turn off training with chat content if you do not want your texts and attachments to be used for model training.
Gemini Data Privacy: What Is Changing Now
Google is splitting a previously combined setting into two independent switches. Until now, the "Keep Activity" option simultaneously regulated how long chats are stored and whether they flow into AI training. Anyone who rejected training could not keep their chats for longer than 72 hours, according to t3n.
Following the update, according to 9to5Google and the Privacy Hub, there are two separate settings:
History: This determines whether and for how long your chats are stored.
Training with chat content: This option separately controls whether texts and attachments from your chats are used to train generative AI models.
Scenario | Previously | After the update |
Keep history, but no AI training | not possible | possible |
Retention period with history disabled | 72 hours | around two months |
Retention period for temporary chats | 72 hours | around two months |
Temporary chats in AI training | no | no |
Automatic deletion with active history | Default 18 months | Default 18 months, selectable 3 or 36 months, or unlimited |
Sources: Gemini Apps Privacy Hub, 9to5Google, t3n
The Catch: Two Months Instead of 72 Hours
If you turn off the history, Google still stores future chats in your account for around two months. During this time, Google uses them to provide responses, evaluate feedback, and secure the services. According to the Privacy Hub, human reviewers may also be involved. One reason Google gives is that the chats should remain available in the event of a system outage. According to the company, Gemini includes the last 24 hours when generating responses.
Google does not explain in detail why a significantly longer storage period is necessary for this. This is also criticized by t3n. Our assessment: Anyone who previously used Gemini deliberately without history should now re-evaluate their usage.
Exception: Feedback
If you submit feedback while history is disabled, for example via the thumbs-up or thumbs-down icon, Google is permitted, according to the Privacy Hub, to use your feedback, the chats from the last 24 hours, and their contents, such as attachments or data from connected apps, to improve Gemini. Reviewed feedback remains stored for up to three years, disconnected from your Google account.
When the Update Arrives
According to the official blog post, Google is rolling out the change worldwide over the coming weeks and informing users via email. If you previously had "Keep Activity" disabled, both new settings will be turned off after the transition (9to5Google). According to Notebookcheck, a German account still showed the old 72-hour setting on November 17, 2024. Incidentally, the separation was a specific user request: In Google's own developer forum, users had demanded this exact separate switch.
Which Data Is Specifically Collected?
Gemini collects significantly more than just the questions you type. According to the Gemini Apps Privacy Hub, this includes, among other things:
What you enter and share: prompts, files, photos, videos, screen content, imported chats, as well as recordings and transcripts from Gemini Live.
What Gemini generates: texts, code, images, videos, music, public links, and summaries.
What happens in the background: reasoning steps of the model, tasks, and tool calls.
Data from connected apps: for instance from Gmail, Drive, or Photos, as well as from third-party services, including tools via the Model Context Protocol (MCP).
Location: the approximate area derived from your device, IP address, or the home and work addresses saved in your Google account.
Before Google saves location data in your history, it coarsens it to an area of more than 3 square kilometers with at least 1,000 users. According to its own statements, Google does not currently use your chats for advertising.
Human Reviewers and the Three-Year Rule
A portion of the chats is read by human reviewers, including trained service providers. Google disconnects the chats from the account beforehand. Important for you: Reviewed chats remain stored for up to three years, even if you delete your history or turn off training later. Google therefore explicitly advises against entering confidential information that you would not want a reviewer to see.
Gemini Data Privacy for Images
When you upload an image, Gemini additionally analyzes it using Google Lens technology and extracts any text it contains. Afterward, the image is treated like any other prompt. If training with chat content is active, photos and files from your chats can also flow into the training of generative models. If you share a photo from a connected gallery such as Google Photos while history is active, Google uses it to improve its services, including with human reviewers.
By contrast, Google does not use audio, videos, and screen shares from Gemini Live by default to improve its services. There is a separate switch for this. Furthermore, only upload content to which you hold the rights, and ask people for permission before recording them in a live chat. Google itself points this out in the Privacy Hub.
Choose These Settings to Better Protect Your Data
With just a few clicks, you can determine what Gemini stores and for what purpose. As soon as the update arrives in your account, we recommend following this order. The menu labels may vary slightly during the transition.
Open the activity page: In the Gemini web app, go to "Settings and help" and then "Activity". Alternatively, you can access it directly via the Gemini Apps Activity page in your Google account (Google Help).
Configure history: Decide whether Gemini visibly saves your chats. When history is active, Google deletes chats after 18 months by default. You can switch to 3 or 36 months, or turn off automatic deletion.
Disable AI training: Turn off the setting for training with chat content if you do not want your texts and attachments used for model training. Future chats will then not be used for this purpose, provided you do not send feedback.
Check audio and Gemini Live: The use of audio, live videos, and screen shares is turned off by default. Check to make sure this remains the case for your account.
Use temporary chats: Start a temporary chat for sensitive topics. It does not flow into training or personalization, though it is also stored for around two months.
Check Memory: In the personalization settings, you can view, edit, and delete what Gemini knows about you. Alternatively, ask Gemini directly: "What do you know about me?"
Clean up connected apps: Disconnect apps that Gemini does not need. For custom third-party connections via MCP servers, Google does not monitor or secure the data, according to the Privacy Hub.
Delete old chats: Delete individual chats or the entire history. Keep in mind: Chats that have already been reviewed remain stored for up to three years.
Important: The Gemini settings only apply to Gemini. Other Google settings, such as Web & App Activity or Timeline, continue to save data independently unless you adjust them separately (Privacy Hub).
Gemini Data Privacy Compared to ChatGPT
With this update, Google catches up with OpenAI regarding training controls, but stores chats without history for longer. A direct comparison of the official statements:
Criterion | Gemini (Personal Account) | ChatGPT (Personal Account) |
Keep history, opt out of training | yes, after the update | yes, via Data Controls |
Retention period for temporary chats | around two months | copy kept up to 30 days for safety reasons |
Temporary chats in training | no | no |
Deleted chats | reviewed chats up to three years, disconnected from account | deleted within 30 days, unless de-identified or kept for security or legal reasons |
Business accounts | no training outside your own domain without permission | Business, Enterprise, and Edu have no training by default |
Sources: Gemini Apps Privacy Hub, Generative AI in Google Workspace Privacy Hub, OpenAI Help on Data Usage
Incidentally, Anthropic also states 30 days for incognito chats in Claude, according to t3n. However, both providers may retain specific data for longer for legal reasons.
One provider is not automatically better when it comes to data privacy. The decisive factors are the account type, the settings, and the question of which data you enter in the first place. With Gemini, the main factors to consider are the longer retention of reviewed chats and the two-month storage period without history.
Gemini's Safeguards for Companies
For companies, one question is paramount: Do your employees work with a personal Google account or with a Google Workspace account? According to the Privacy Hub, different terms apply to work and school accounts, which Google describes in the Generative AI in Google Workspace Privacy Hub:
No training without permission: Without consent, content is neither reviewed by humans nor used outside your own domain to train generative models.
Contractual basis: Prompts are considered Customer Data under the Cloud Data Processing Addendum.
Admins control storage: In the Gemini app, administrators determine whether and for how long chats are stored (3, 18, or 36 months, default 18 months).
Feedback without training: Feedback from Workspace does not flow into model training and is retained for up to 18 months.
Assistance with DPIA: Google provides a dedicated guide for the Data Protection Impact Assessment.
Notable: The Workspace Privacy Hub still states up to 72 hours for chats when history is turned off. The new two-month rule so far applies to the Gemini apps for personal accounts. In addition, according to Google, Gemini in Chrome and Gemini Notebook currently do not support ISO, SOC, or FedRAMP compliance.
Business data, especially customer or employee data, does not belong in a personal Gemini account. Consumer rules apply there, including potential human review. Clarify the deployment in advance with your data protection officer and establish binding rules in an AI policy.
Security in Gemini Pro and Google AI Pro
Many people ask whether a paid subscription provides more data privacy. The Gemini Apps Privacy Hub does not provide separate data privacy rules for subscribers. It merely lists subscription information as an additional data category. Google only describes differing conditions for work and school accounts. Therefore, a Google AI Pro subscription on a personal account does not replace a Workspace agreement. In our view, the same applies to choosing a more powerful model: The account and settings are decisive for data privacy, not the model.
Gemini Data Privacy in the EU: GDPR, Rights, and Oversight
In the European Economic Area and Switzerland, your contractual partner is not Google LLC, but Google Ireland Limited. For processing, Google cites four legal bases under EU data protection law in the Privacy Hub:
Performance of a contract: to respond to your requests and provide features.
Legitimate interests: among other things, to improve models, train new models, and prevent abuse.
Legal obligations: for example, in response to requests from authorities.
Consent: for features such as Voice Match, Memory, and the use of audio and live recordings.
This gives rise to specific rights. You can object to the processing of your data and request the correction of inaccurate personal information in Gemini responses. This can be done via a form in Google Help or directly in the app via "More" and "Report a legal issue". You can withdraw any consent you have given at any time.
The Irish Data Protection Commission (DPC) is the lead supervisory authority for Google. How closely it looks was demonstrated by its decision on September 21, 2022: a fine of 403 million euros due to the processing of location data in the Web & App Activity, Location History, and Location Accuracy features (EDPB). The case did not concern Gemini, but rather the period from 2018 to 2020. However, since Gemini also processes location data, it is worth taking a close look at these settings.
Conclusion: More Control, But Not a Free Pass
Google finally gives you the choice to keep your history and opt out of AI training. In return, Google stores chats without history for significantly longer, and reviewed chats remain stored for up to three years. Check your settings as soon as the update arrives, and as a general rule, do not enter confidential data into AI chats.
Would you like to use AI tools like Gemini safely in marketing? We support your team with practical AI expertise and a non-binding initial consultation!
* Note: This article is for general information purposes only and does not replace legal advice. For a binding assessment of your AI deployment, please contact your data protection officer. Information is subject to change at any time.
FAQ on Gemini Data Privacy
Can Google read my Gemini chats even if history is disabled?
Yes. Even without history, chats remain stored for around two months and can be viewed by human reviewers for safety purposes. Google disconnects them from your account beforehand. Reviewed chats remain stored for up to three years.
Are Gemini chats still saved if training is turned off?
Yes. Training and storage are separate settings. With active history, chats remain stored for 18 months by default, and for around two months without history. Even without training, Google uses chats to generate responses and keep its services safe.
Is Gemini more privacy-friendly with a paid subscription?
Not automatically. The Privacy Hub does not provide separate data privacy rules for subscribers. Differing conditions only apply to work and school accounts via Google Workspace.
May I enter confidential company data into Gemini?
Not in a personal account: Google itself advises against entering confidential information there. With Google Workspace, content is neither reviewed by humans nor used for training outside your domain without permission. Nevertheless, coordinate the usage in advance with your data protection officer and, despite these assurances, do not assume that the data is completely protected.
Sources
Google Blog: We’re updating controls in the Gemini app (16.11.2024)
Google: Gemini Apps Privacy Hub (as of 17.11.2024)
Google: Generative AI in Google Workspace Privacy Hub (as of 06.10.2024, page updated 10.11.2024)
t3n: Mehr Datenschutz bei Gemini: Nutzer können jetzt KI-Training untersagen, aber Verlauf behalten (18.11.2024)
9to5Google: New Gemini setting lets you keep chat history, but turn off AI training (16.11.2024)
Notebookcheck: Gemini separates history from AI training but keeps chats longer (17.11.2024)
OpenAI Help Center: How OpenAI handles data in consumer services
EDPB: The Irish Data Protection Commission fines Google EUR 403 000 000 (decision from 21.09.2022)




