Google made Deep Review public on October 14. The agent checks technical documents and provides feedback on consistency and content; the launch expands a scientific-author assistance tool toward other knowledge documents, including legal and financial material. It initially accepts PDF and plain text.
The review pipeline segments the document, assigns compute budgets, runs verification agents in parallel and synthesizes their findings with consistency and source checks. Multi-file input is still planned, rather than part of the initial PDF and plain-text offering.
Each user receives two free reviews per week. A personal Gemini API key enables higher limits and programmatic integration. Google describes a four-stage pipeline and cites over 16,000 manuscripts reviewed in opt-in conference-related programs: private author feedback does not replace an official peer-review decision.
The data policy has two parts: documents are not used to train, fine-tune or improve Google models, but they are retained for one year for display in the interface. Confidential material therefore needs an internal policy check before upload, even without training use.