Gemini API now stores requests by default in new projects

Google changed the Gemini API's Generate Content API on October 26, 2026: new paid-tier projects now store generateContent, streamGenerateContent and batchGenerateContent requests by default (store=true), and keep them as logs for up to 55 days. Projects created before that date keep their current setting, which is off unless someone turned logging on in Google AI Studio. If your team creates Gemini projects for new apps, environments or customers, decide now whether each one should keep prompts and responses, and for how long.

The catch is that two projects running the same code can now behave differently. An app moved into a project created this week will store full prompts and responses unless the project setting or the request says otherwise, while the same app in last month's project stores nothing. Google's Logs and datasets guide says the request-level store property "takes precedence over the project-level setting," so setting it in code is the way to make storage independent of when a project was created.

What changed in Gemini API logging?

New projects now log Generate Content API calls by default, the same way the Interactions API already did. The Gemini API release notes say new paid-tier projects "now store requests by default (store=true), matching the Interactions API. Existing projects keep their current setting."

Scroll sideways to see all 3 columns.

Comparison dimension	Projects created before Oct. 26, 2026	Projects created on or after Oct. 26, 2026
generateContent and streaming/batch variants	Not stored (store=false) unless logging was turned on in AI Studio	Stored (store=true)
Interactions API	Stored by default	Stored by default
Per-request store value	Overrides the project setting	Overrides the project setting

Logging covers calls made through Google's OpenAI compatibility endpoints too, according to the same guide. Storage is available only on the paid tier, so free-tier projects are not part of this change.

What a stored request contains and where it goes

A stored request is a log entry you can open in the Logs page of Google AI Studio. Google says you can "inspect the full prompt and response from Gemini, and the context from the previous turns." For Interactions API calls, the entry also links to the previous interaction.

Google's data logging and sharing page describes these logs as "private to your Google Cloud project" and separate from logs kept for abuse monitoring. It also says that, by default, prompts and responses in these logs "are not used for product improvement or development." That changes only if someone saves logs to a dataset and shares it with Google. Shared datasets fall under the terms for unpaid services, which allow model training and say "human reviewers may read, annotate, and process" what you share.

So the new default does not send your data to model training. It does keep a readable copy of every supported prompt and response inside the project, which matters if those prompts carry customer records, contract text or support conversations.

How long Gemini API logs are kept

Logs expire after 55 days by default. A project can shorten that window to 7, 14, 28 or 55 days. Logs saved to a dataset have no retention period and stay until someone deletes them, and datasets can be exported as CSV, JSONL or to Google Sheets.

Some calls aren't logged at all. The guide lists Imagen and Veo models, Gemini embedding models, the Gemini Robotics model, inputs containing videos, GIFs or PDFs, and public preview agents as unsupported for logging.

One operational detail is easy to miss: a project needs at least one active API key for its logs to show. Google notes that if every key is deleted, "for example, if a leaked key was deleted," you lose visibility of the logs until a key exists again. Google's wording describes lost visibility, not deletion, so treat key deletion as a separate step from clearing logs.

How do you turn off Gemini API logging?

There are two controls. At the project level, a project owner can switch Generate Content API logging and Interactions API logging on or off separately in the Settings panel of the Logs dashboard in AI Studio. At the request level, you set store to false (or true) in the request config.

The per-request setting for Generate Content needs a recent SDK. Google lists these minimums:

• Python: google-genai 2.33.0 or later
• JavaScript: @google/genai 2.31.0 or later
• Java: google-genai 1.79.0 or later
• Go: google.golang.org/genai 1.76.0 or later

Be careful with the Interactions API toggle. Google warns that turning Interactions logging off at the project level stops the API from "automatically storing and retrieving conversation history unless explicitly overridden per-request." The BaristaLabs explainer on how the Interactions API stores state covers why multi-turn chaining depends on storage.

Who does this affect?

It affects any team that creates a new paid-tier Gemini project from now on. Common cases are a separate project per environment, per client or per product, a fresh project for a pilot, and internal tools that someone sets up in AI Studio. Existing production projects don't change unless someone changes them.

It also affects teams whose Gemini calls go through a shared client wrapper. If that wrapper never sets store, the storage behavior now depends on which project the API key belongs to, not on the code. The same kind of drift showed up in the Gemini 3.6 Flash request settings change, where old payload fields kept traveling through wrappers long after anyone remembered them.

Why it matters for teams handling customer data

BaristaLabs interpretation: the new default is useful for development. Stored logs give you a debugging trail and a way to build evaluation sets from real traffic, which teams otherwise have to build themselves. The cost is a second copy of prompt data, held for up to 55 days in a place your data-retention policy may not list. If you promised customers a retention period, or if your prompts include personal data, that copy needs to be on purpose. We made a similar argument about keeping agent logs private: logs are worth keeping, but only with a deliberate owner and a deliberate retention period.

A first step: make storage explicit in code

BaristaLabs recommends setting store explicitly in your Gemini client code instead of relying on project defaults:

1. List the Gemini projects created on or after October 26, 2026, and check the Generate Content logging toggle in each one.
2. Upgrade to the minimum SDK version for your language, then set store in the shared client: false for production paths that carry customer data, true where you want logs for debugging or evaluation.
3. In projects that keep logging on, set the shortest retention window that still supports debugging, such as 7 or 14 days.
4. Decide who may create datasets, since datasets don't expire, and who may share them with Google.
5. For calls that go through the OpenAI compatibility endpoints, use the project-level toggle. The guide's request-level examples cover Google's own SDKs only.

If you're also reviewing how API keys are scoped and rotated, the Gemini API key leak pattern is worth reading next. For help setting a retention policy across AI providers, see BaristaLabs AI consulting.

Sources and scope

• Google AI for Developers, Gemini API release notes, entry dated October 26, 2026.
• Google AI for Developers, Logs and datasets, last updated October 26, 2026, retrieved October 27, 2026.
• Google AI for Developers, Data logging and sharing, last updated October 26, 2026, retrieved October 27, 2026.

These pages don't say who inside an organization can view a project's logs, and they don't describe a way to set store on OpenAI compatibility calls. BaristaLabs hasn't tested the new default in a live project; everything above is from Google's documentation.