Skip to content
AI.info

The Pulse

Google makes Gemini classification generally available in Drive

Google Workspace administrators can use natural-language instructions to classify Drive files with Gemini. The release adds instruction testing and reporting, with a rollout expected to finish by December 15.

Google makes Gemini classification generally available in Drive

Google Workspace administrators can now use natural-language instructions to classify files in Drive with Gemini, as the feature moves into general availability. The release follows an open beta and adds instruction testing and reporting, according to a Google Workspace Updates post.

The rollout for Rapid Release and Scheduled Release domains began on November 17, 2026, and will take more than 15 days for feature visibility. Google expects it to finish by December 15.

Instructions replace training files

Previously, administrators had to identify and manually label training files so an AI model could learn the data associated with each classification level. The Gemini-based approach offers an alternative: administrators write instructions instead of manually training a model.

Administrators select a label, provide instructions and define the audience for the files to be evaluated. Gemini interprets those instructions, evaluates files and applies classification labels.

Automated labels can be used with data loss prevention policies, retention rules and audit investigations. Google says the approach can help organizations enforce those policies across Drive.

Testing and reporting

Before deployment, administrators can test and refine instructions in the Admin console to preview which labels Gemini would apply. A model details page provides near-real-time metrics, including the number of labeled files, user acceptance of labels and Drive inventory reporting.

File owners and editors with the appropriate label permissions can review and accept or modify a Gemini-applied label. Audit logs record labeling activity, including acceptance or changes by users.

Organizations can have one active Gemini instruction model and create up to five AI classification models in total, combining Gemini instruction models and custom-trained machine learning models.

File and label limits

Files must be in Shared Drives or owned by users with supported licenses. The Gemini model typically needs at least 50 words and 200 characters of readable or OCR-extracted text for a high-quality verdict. Client-side encrypted files and non-text media without readable text are skipped.

The labels must use a published single-select options list or badge list field. Gemini classification will not overwrite labels that users applied manually, accepted or modified, or labels applied by data loss prevention rules.

Enabling auto-apply starts a one-time scan of existing and inactive files. The scan typically starts within one to two weeks, but completion may take weeks or months depending on the size and volume of files.

Updated instructions apply to future uploads and new content. On-demand rescanning of previously evaluated files is not included in this release.

Access and data regions

Google says eligible Drive content is processed and stored in an organization's designated US or Europe data region, based on the file owner or Shared Drive policy. The company says Drive files and prompts remain isolated to the organization and are not used to train its global models.

The feature is available for Enterprise Plus, Frontline Plus and Google AI Pro for Education. Administrators with a supported license can configure instructions under Data classification in the Admin console's Security section; there is no end-user setting.

Source

  1. Google Workspace Updates

Companies in this story

Related stories

Explore

More articles