Deep Review

Google, Google DeepMind, Springer NatureBuilt by: big techStatus: LiveYear: 2026

Last updated October 2026

Overview

Google Research’s Deep Review checks research papers and other technical documents for errors, verifying theoretical results and experiments and flagging potential flaws. It grew out of the Paper Assistant Tool (PAT), which authors could use to check computer science papers privately before submitting them to STOC 2026, ICML 2026, NeurIPS 2026 and ICLR 2027. Google released Deep Review worldwide on 14 October 2026, and says it can also check other documents, such as design documents, presentations, and legal and financial documents.

Specialist AI agents check sections of a document in parallel, with more effort given to dense sections such as proofs. A final agent combines their reports and checks claims against external sources.

Benefits and outcomes

  • Google reports that authors submitted more than 16,000 papers to the PAT pilots.
  • ICML reports that about 4,500 ICML 2026 papers received PAT feedback.
  • In ICML’s survey, with 869 responses, 35.4% of authors of papers containing theory said PAT found significant gaps in the theory that took more than an hour to fix.
  • In the same survey, 31% of authors of papers with experimental results ran new experiments based on the feedback.
  • In the same survey, 92.1% said they would use the tool again.
  • GIGAZINE and DigitalToday report that Google’s paper on PAT, the earlier system, found 89.7% of 29 known errors in 26 mathematics and computer science papers, compared with 55.2% for Gemini 3.1 Pro used on its own.

Notes and quotes

  • “Deep Review handles some of the most cognitively burdensome facets of review, while leaving higher-level judgements to humans.” Google Research.
  • “Documents uploaded to Deep Review are never used to train, fine-tune, or improve Google models.” Google Research, which says uploads are stored for one year for display to the user.
  • DigitalToday reports Google’s acknowledgement that Deep Review can flag valid proofs as errors.
  • On 26 October 2026 Springer Nature and Google DeepMind announced a research partnership to co-design an AI peer review assistant using Gemini models, building on Google’s ScholarPeer and Deep Review. The assistant helps reviewers work through the text, charts and figures of a manuscript and find the parts that need closer examination. A prototype is being tested in Springer Nature’s private Gemini Enterprise environment, so manuscripts stay within Springer Nature’s systems.

Sources

This entry was written with the help of AI, from the sources listed. Please check the sources before relying on it, and let us know if anything needs correcting.