Google Research has publicly released Deep Review, an agentic AI system for rigorous feedback and technical verification on complex documents. The tool evolved from the Paper Assistant Tool (PAT), which was piloted with four flagship computer science conferences—STOC 2026, ICML 2026, NeurIPS 2026, and ICLR 2027—and provided pre-submission feedback to authors of more than 16,000 manuscripts who opted in.
The system is designed to address what Google calls a "validation bottleneck": AI accelerates content creation, but human capacity for verification has not kept pace. Deep Review checks logical consistency and mathematical statements, suggests improvements, grounds facts, and compares with existing work. It now extends beyond academic papers to legal and finance documents, and can be used both during drafting and as part of formal review. Google cites the pressure on peer review, noting over 74,000 submissions to ICLR, NeurIPS, and ICML in 2026, up from roughly 23,000 in 2023.
Public access includes two complimentary document reviews per week, with higher usage limits and programmatic integration available through a Gemini API key. Input formats are currently PDFs and plain text, with multi-file support planned. Google states that uploaded documents are never used to train, fine-tune, or improve its models, and are stored for one year solely for display in the Deep Review UI. The system uses a four-stage agentic pipeline beginning with autonomous segmentation of documents into logical segments, a design intended to overcome the failure of single-prompt LLM calls on long, nuanced inputs.