If you created a paid-tier Gemini API project on or after October 26, 2026, Google is keeping a copy of your generateContent, streamGenerateContent and batchGenerateContent requests for 55 days unless you say otherwise. The changelog entry for that date, “Generate Content API storage on by default”, says new projects using the Generate Content API on the paid tier now store requests by default (store=true), matching the Interactions API, and that “Existing projects keep their current setting.” Gemini API changelog · Logs and datasets
Older projects are untouched: their Generate Content logging stays off unless someone turned it on in AI Studio. Nothing in the docs says they will be migrated, and nothing addresses what happens when an old free-tier project upgrades to paid, so treat that case as unknown and check the Settings panel after upgrading. The Interactions API already stored by default, so teams on it see nothing new, only the same behavior extended to Generate Content. Logs and datasets
The log sits in your own project’s dashboard, and Google’s policy page says logged prompts and responses are not used for training unless you share a dataset. The two mistakes to avoid run in opposite directions: assuming the default means Google trains on your prompts, and assuming store=false means Google holds nothing. Abuse-monitoring logs, grounding data, Live API session state, uploaded files and explicit caches are all outside this switch.
What gets logged, and what never does
Logging applies to “All GenerateContent, BatchGenerateContent, StreamGenerateContent API calls”, to Interactions API calls other than Managed Agents, and “includes calls made through OpenAI compatibility endpoints”. It is available only to projects on the paid tier, which the policy page phrases as billing-enabled projects. Logs and datasets · Logs policy
| Surface | Logged by the project setting? |
|---|---|
generateContent, streamGenerateContent, batchGenerateContent | Yes, default on for projects created from Oct 26 |
| Interactions API (except Managed Agents) | Yes, default on already |
| OpenAI-compatible endpoints | Yes |
| Imagen and Veo models | No |
| Gemini embedding models | No |
| Gemini Robotics model | No |
| Inputs containing videos, GIFs or PDFs | No |
| Public Preview Agents | No |
Source: Logs and datasets. The page does not say whether cached content, uploaded files or grounding context appear inside a logged request beyond these exclusions, and we could not find a statement on how store is set through the OpenAI-compatible endpoints. That a request with a PDF attached is not logged says something about this log only; the PDF itself sits in the File API under its own retention, covered below.
Turning it off: the request beats the project
“The request-level store property takes precedence over the project-level setting.” So a request carrying store=false is not logged even when the project has logging on. That is the control to reach for in code that handles anything sensitive, because it survives whatever a colleague later does in the dashboard. The docs do not give the reverse example (a store=true request in a project with logging off), so we will not assert it. Logs and datasets
The project-level switch lives in the Settings panel of the Logs dashboard in AI Studio, with separate toggles for the generateContent API and the Interactions API. The policy page says a project owner can turn logging on or off at any time from the Logs and Datasets page. One warning from the docs matters if you use both: turning Interactions API logging off in that panel stops the API from automatically storing and retrieving conversation history unless a request overrides it. The Generate Content toggle has no such side effect, so do not flip both reflexively. Logs and datasets · Logs policy
In the SDKs, store for Generate Content calls sits inside the request config, not at the top level (the Interactions API takes it as a top-level argument on the create call). Google’s examples all pass True with the comment “Set to False to disable logging of this request”; the opt-out below is that example with the value flipped, which we have not run. Logs and datasets
from google import genai
client = genai.Client()
response = client.models.generate_content(
model="gemini-4-argon",
contents="Summarise this contract clause.",
config={"store": False}, # the docs' example passes True
)Sending store at all needs a recent SDK. The page lists minimums for four languages and says nothing about .NET, raw REST or what an older SDK does with the field. Until a team is on these versions, the project toggle is likely the only opt-out it has, though the docs do not say so in those words. Logs and datasets
| SDK | Minimum to set store on Generate Content |
|---|---|
Python google-genai | 2.33.0 |
JavaScript @google/genai | 2.31.0 |
Java google-genai | 1.79.0 |
Go google.golang.org/genai | 1.76.0 |
Retention: 7 to 55 days, or forever in a dataset
Logs “will expire and be marked for deletion after a default retention window of 55 days”. AI Studio lets you shorten that to 7, 14 or 28 days; 55 is the maximum. Logs saved to a dataset do not expire, and the policy page says datasets have no set retention period. Logs and datasets · Logs policy
| Setting | Retention |
|---|---|
| Default | 55 days |
| Options | 7, 14, 28 or 55 days |
| Saved to a dataset | No expiry |
A dataset is therefore the only part of this feature that outlives the window, and it only exists if someone creates one. A compliance review should ask whether anyone has saved a dataset, since “is logging on” misses it: a 7-day retention setting says nothing about rows that were copied out before they expired.
Training: not from logs, yes from shared datasets
The policy page is plain about the default: “prompts and responses within logs are not used for product improvement or development”, a position it ties to logging being limited to billing-enabled projects. Logs policy
Sharing a dataset with Google is the opt-in that changes it. Shared data “may be used to improve model quality, and inform the training and evaluation of future models and services”, and “human reviewers may read, annotate, and process the API inputs and outputs you share”. Google says it disconnects that data from your Google Account, API key and Cloud project before reviewers see it. There is also a Contributor Tier, invite-only and in private preview, that offers up to 50% off token costs on select models in exchange for permission to use your data. Logs policy
So the default-on change moves no one closer to training. It does mean the data that could later be shared now exists by default, for 55 days, in a dashboard, which matters to a team with contractual limits on where customer prompts may sit.
What store=false leaves in place
The policy page says a project’s logs are “separate from any logs held solely for abuse monitoring purposes” and says nothing about how those are handled. The zero-data-retention page does: “for Paid Services, Google logs prompts and responses for a limited period of time solely for detecting violations”. No period is given. Logs policy · Gemini API, zero data retention
The same page lists the features that keep data no matter how store is set. Gemini API, zero data retention
| Feature | What is kept | How long | Can you avoid it? |
|---|---|---|---|
| Abuse monitoring (paid services) | Prompts and responses | “a limited period of time”, unstated | No |
| Grounding with Google Search | Prompts, contextual information, generated output | 30 days | No, if you use the tool |
| Grounding with Google Maps | Same, used for reliability engineering | 30 days | No, if you use the tool |
| Live API | Conversation state (text, audio, video) | Up to 24 hours if a session handle is generated | Do not configure SessionResumptionConfig |
| File API | Uploaded files at rest | Until deleted or expired | Delete files yourself |
| Explicit context caching | The cached context | Your ttl or expire_time | Do not use cached_content |
| Implicit in-memory caching | Request data in RAM | 24-hour TTL | Google says this “does not violate zero data retention” |
For regulated workloads the page is direct: “If your workload requires guaranteed zero data retention or enterprise data processing agreements, use Vertex AI.” Google links a zero data retention guide for the Gemini Enterprise Agent Platform, which loads, and sets out which features there retain data and which settings control it. The Gemini API itself is not where Google offers a ZDR guarantee. Gemini API, zero data retention
Who should act
Anyone who created a paid project since October 26. Open the Logs dashboard and look. If the log should not exist, turn off the Generate Content toggle and add store=false to the calls that matter, in that order; the request override is the one that holds.
Teams with data-handling clauses that name retention periods. Decide whether 55 days of logs on Google’s side fits the contract. If not, shorten the window, or switch it off, and then check whether a dataset was ever created, because that is the part with no expiry.
Teams on older SDKs. Python below 2.33.0, JavaScript below 2.31.0, Java below 1.79.0 or Go below 1.76.0 cannot send store. Upgrade before relying on the request-level override; use the project toggle in the meantime.
Teams that were already on the Interactions API. Nothing changed for you on October 26. Just do not turn off Interactions logging to tidy up, because that also stops automatic conversation history.
Anyone who needs zero retention. The toggle is not that. Google’s own answer is Vertex AI, and the ZDR page’s table of what is kept regardless of store is the list to put in front of whoever signs off on the vendor review.
Our judgment: the default is defensible for what it is, a debugging and evaluation log with a hard expiry, no training use without opt-in, and a request-level off switch that beats the dashboard. The problem is that it arrived as a changelog line, in a setting most teams never open, and store=false reads like a full stop when it leaves the retention in the table above running. Read the ZDR page before the logs page, and check the Settings panel when you create a project rather than discovering it during an audit.
This follows our Gemini 4 Argon pricing and migration notes and the October 8 Flash rerouting and Deep Research shutdown, and like those changes it appeared first as a changelog entry. The same changelog marks Computer Use as generally available on October 30; our Computer Use checklist covers the screenshot-logging angle.
Sources
Google: Gemini API changelog · Logs and datasets · Logs and datasets data policy · Gemini API and zero data retention · Gemini Enterprise Agent Platform zero data retention guide
Related S5 Labs coverage: Gemini 4 Argon pricing and migration · Gemini API Flash rerouting and Deep Research shutdown · Gemini API Veo and TTS preview shutdowns