Google Launches Deep Review AI Tool for Error Detection

Google released Deep Review, an AI tool that detects logical contradictions and mathematical errors in research papers and technical documents, to general users worldwide on October 14, 2026, according to digitaltoday.co.kr.

Deep Review splits long documents into sections such as introduction, methodology, results and mathematical proofs, assigning AI agents to each area in parallel. A separate agent then consolidates the results and cross-checks them against external material to reduce duplicate or incorrect findings.

In testing with the existing Paper Assistant Tool on 26 math and computer science papers, Deep Review found 89.7% of 29 confirmed errors. Single use of Gemini 3.1 Pro achieved a detection rate of 55.2%, about 34.5 percentage points lower.

General release details

The general release adds support for the latest AI model Gemini 4 Argon. Deep Review can also check inconsistencies in software design documents, legal contracts, financial statements and presentations.

The free usage limit is twice per week, expandable by connecting a Gemini API key. Supported input formats are initially PDF and plain text.

In October 2025, Google conducted trial operations with major conferences including STOC 2026, ICML 2026, NeurIPS 2026 and ICLR 2027 on over 16,000 research papers. In an ICML 2026 trial survey, 35.4% of respondents said they received reports of major theoretical problems, and 31% said they conducted new experiments reflecting the AI's findings.