An AI system built to scrutinize research papers is opening to a wider range of technical documents. Google released Deep Review publicly on October 14, following an earlier tool that gave authors feedback before conference submission.
The point is to test a document's reasoning, rather than merely shorten it. Google says the system checks mathematical statements, internal consistency and supporting evidence. That is a different task from producing a fluent summary.
Deep Review divides a document into logical sections, which can overlap. It allocates more computing effort to demanding material, sends sections to specialized reviewing agents and combines their reports. A final stage checks consistency across sections and consults outside sources.
Each reviewing agent can see the full document, according to Google. That matters when a claim in one section depends on a definition or experiment elsewhere.
The earlier Paper Assistant Tool provided optional feedback on more than 16,000 manuscripts, Google says. That was separate from official conference decisions. Participation and author approval are not measurements proving that every criticism was correct.
The public version initially accepts PDFs and plain text. Google offers two complimentary reviews per week, with higher-volume access through a connected Gemini API key.
The data policy also deserves attention: Google says uploaded documents are not used to train its models, but are stored for one year for display in the interface. Those are two separate promises.
An automated critique still needs checking against the document and its sources. Deep Review produces feedback, not a publication decision.
