Google launches Deep Review AI tool to help researchers catch manuscript errors

Google Research on Oct. 14 released Deep Review, an AI system designed to flag mathematical errors, logical gaps and unsupported claims in technical documents. The public launch follows pilots that provided feedback on more than 15,000 manuscripts, according to Google.

Speaking at Google’s Research@ event in Mountain View, California, Rajesh Jayaram, a research scientist at Google Research, said AI can now generate drafts of academic papers and legal documents “nearly at the push of a button.” Human capacity to verify that output, he said, has stayed essentially fixed.

“We need to make sure that all this content we’re generating is correct, that it’s reliable, and that it makes sense in the context that it’s being used,” he said.

Google’s new agentic AI system Deep Review is intended to close that verification gap. The system grew out of Google’s Paper Assistant Tool (PAT), which offered authors private, optional feedback before submission deadlines at four computer science conferences: the Symposium on Theory of Computing (STOC), the International Conference on Machine Learning (ICML), NeurIPS and the International Conference on Learning Representations (ICLR). The pilots operated separately from formal peer review.

Nika Haghtalab, a UC Berkeley computer scientist and NeurIPS program chair, said AI has also made it easier to produce papers “that look very polished, but unfortunately have weak substance.” Providing feedback before submission can help authors address weaknesses while preserving reviewers’ time, she said.

Deep Review Works by dividing documents into sections and allocates more computing resources to difficult material. Users can also specify the overall level of model effort when using Deep Review. After uploading a document, AI agents examine its sections, and a synthesis agent combines their findings and uses Google Search to check for potential hallucinations.

Jayaram said a review takes about 10 to 20 minutes and returns major and minor issues, suggestions and summaries. “The comments will be anchored directly to the text where they talk about it,” he said.

At Google’s Research@, Haghtalab said Deep Review’s detailed verification could give researchers more time to assess “what is novel, what matters, and what kind of work actually moves the field forward.”

In a June preprint, Google researchers reported that the Paper Assistant Tool detected 89.7% of known mathematical and proof errors in a selected benchmark subset, compared with 55.2% for a single Gemini 3.1 Pro response. The evaluation covered 29 errors across 26 papers. Pilot participants also reported cases in which the system incorrectly challenged proofs or arguments, underscoring the need for researchers to check its critiques.

Deep Review accepts a range of file formats, including PDFs, PowerPoint presentations, Word documents, Markdown and plain text. Google offers two complimentary reviews per week using 4 Argon on Medium compute, with higher usage limits available through a connected Gemini API key. The company says uploaded documents are excluded from model training and retained for one year so users can access them through the interface.