Since October 26, 2026, new projects on the paid Gemini API tier have stored Generate Content API requests by default (store=true). Projects created before that date retain their previous setting. Developers can disable project logging in Google AI Studio or pass store=false on a supported request. Google confirmed the change in its release notes, and it does not mean every prompt will automatically be used to train models.
The change concerns the developer API, not the consumer Gemini app. It is also important to distinguish the Generate Content API from the Interactions API, which already used storage by default. Actual behavior depends on the project, API method, service tier, and any request-level override.
Who is affected: API, date, and settings
| API or project | Documented storage default | What to check |
|---|---|---|
| Generate Content, paid project created on or after Oct. 26, 2026 | On, store=true | Disable it in the dashboard or pass store=false per request |
| Generate Content, project created before Oct. 26, 2026 | Retains its prior setting, normally store=false unless previously enabled | Check the dashboard instead of assuming an older project is off |
| Interactions API | On by default, with server-side state storage | Review store=false and conversation-history dependencies |
| Free tier | The new Generate Content logging default was announced for paid projects | Do not assume the paid-tier rule applies; check the API and terms |
The change covers generateContent, streamGenerateContent, and batchGenerateContent. The logs documentation also supports calls made through OpenAI-compatible endpoints and excludes managed agents from the supported logging calls. The announcement does not establish a universal change across every API, project, and plan.
The creation date matters: projects established before October 26 were not collectively switched to store=true. They retain their previous choice. An older project with logging already enabled can keep logging, while one that never enabled it keeps its previous store=false default.
How to disable request logging
To change a project default, open Logs and Datasets in Google AI Studio, select the appropriate project, and open Settings. Turn off storage for the Generate Content API or the Interactions API separately, depending on which service you use. The documentation says the choice can be changed at any time; it does not claim that turning it off immediately erases existing records.
To control one request, set store=false. A request-level setting overrides the project setting. In Python, the documented models.generate_content configuration places the flag inside config, such as config={"store": False}. For the equivalent JavaScript operation, it appears as config: { store: false }. In the Interactions API, the Python example passes store=False directly to client.interactions.create. Parameter placement therefore depends on the method and language.
For Generate Content requests, Google documents minimum SDK versions for the store field: google-genai 2.33.0 for Python, @google/genai 2.31.0 for JavaScript, google-genai 1.79.0 for Java, and google.golang.org/genai 1.76.0 for Go. Check your dependencies and the method signature before adapting older code. These examples were checked against documentation, not executed against a production project.
With the Interactions API, turning storage off prevents automatic server-side conversation-history retrieval unless a request enables storage again. The API documentation also says store=false is incompatible with background execution and continuing conversations through previous_interaction_id. This is a functional trade-off, not merely a dashboard display preference.
What is logged, and for how long
For eligible paid projects, developers can inspect full prompts, model responses, and context from earlier turns in AI Studio. The guide also notes that at least one active API key must remain in a project for logs to be visible. Deleting every key can hide the records from view, but is not documented as a way to delete logs.
The default log-retention period is up to 55 days, after which logs are marked for deletion. Project owners can choose 7, 14, 28, or 55 days. Logs moved into datasets have no fixed expiry and require separate management. These periods describe the documented project logs, not every category of technical data a service might process.
Google currently excludes Imagen and Veo, embedding and robotics models, inputs containing video, GIF, or PDF files, and public-preview agents from this logging system. These exclusions define the logs feature; they do not establish that other processing or safety-related records never occur.
Storage is not the same as training
Google distinguishes three activities: keeping private developer-owned logs within a project, voluntarily sharing examples with Google, and holding separate records solely for abuse monitoring. Under its Gemini API logging policy, prompts and responses in paid-project logs are not used for product improvement or development by default. The new store=true setting is therefore not automatic authorization to train models.
Voluntarily shared datasets and feedback follow different rules. Google says contributed datasets can be used to develop and improve systems, including model training, and warns against sharing personal, confidential, or sensitive content. The policy also describes an invitation-only Contributor tier with its own data-use conditions. Turning project logging off should not be represented as proof that all other recording ends or that these separate data-sharing terms change.
What to verify before releasing an application
For any new or existing integration, identify the creation date of the billed project, the API method actually called, whether the SDK supports store, the Logs and Datasets setting, and whether a request overrides it. Check the retention period and any datasets stored separately. For systems processing user information, review privacy notices and whether identifiable details need to be sent at all before keeping logs.
Brazil appears on Google’s official list of regions supported by the Gemini API and Google AI Studio. This does not identify where each item of data is physically processed or confirm the configuration of a particular Brazilian project. This is a documentation-based analysis checked on October 29, 2026, without account access, live API calls, or hands-on verification of any project’s logging state.
