Google Research announced on October 14, 2026 the public launch of Deep Review, an agentic AI system designed to provide rigorous feedback and technical verification on complex documents. The tool evolves from the Paper Assistant Tool (PAT), which had already been tested on more than 16,000 manuscripts at conferences such as STOC, ICML, NeurIPS and ICLR.
Each user receives two free analyses per week. For higher volumes, it is possible to connect a Gemini API key. The system initially accepts PDFs and plain text, and uploaded documents are not used to train Google models.
How it works
Deep Review operates in four stages: document segmentation, adaptive allocation of computational resources according to information density, parallel verification by specialized agents, and global synthesis with external checks to avoid hallucinations. The goal is to alleviate the validation bottleneck at a time when the production of technical and scientific content is growing rapidly with the help of AI.
In earlier PAT tests, the tool detected about 89.7% of known mathematical and proof errors in a subset of papers, above the performance of a simple model call.
Sources
- Google Research: Analyzing technical documents with Deep Review
- arXiv: Towards Automating Scientific Review with Google’s Paper Assistant Tool
Transparency: This content was created, edited or reviewed with the assistance of artificial intelligence. The information was cross-checked with public posts on X and sources available on the internet. Consult the original sources to verify the full context.
By GeekikiBot
