# Google’s Deep Review checks papers for errors, and now anyone can use it

URL: https://ai-pulse.aglaganova.com/blog/googles-deep-review-checks-papers-for-errors-and-now-anyone-can-use-it
Published: 2026-10-14
Author: Roman Aglaganov, Full-stack developer
Category: Research
Region: Global
Tags: science, reasoning, agents

> Tested on more than 16,000 manuscripts at four major computer science conferences, Google’s AI reviewer is now public. It checks proofs, logic and claims, and leaves the final verdict to people.

![Illustration: four pages of text, each with a magnifying glass, one line on the third page marked in red](https://ai-pulse.aglaganova.com/api/media/file/review-1.png)

AI has made writing papers faster. Checking them has not got any faster. Three big AI conferences, ICLR, NeurIPS and ICML, received more than 74,000 submissions this year, up from about 23,000 in 2023, and each one needs several expert reviews. On 14 October, Google launched Deep Review, an AI system for checking long technical documents, to the public.

## Where it comes from

Deep Review grew out of Google’s Paper Assistant Tool. Over the past year, authors of more than 16,000 manuscripts at STOC 2026, ICML 2026, NeurIPS 2026 and ICLR 2027 opted in to get private feedback from it before submitting. That feedback was kept separate from the official peer review, and Google says the papers were not used for training.

## How it works

Asking a single chatbot to review a 40-page paper tends to fail: it runs out of reasoning budget or misses links between sections. Deep Review splits the job up:

**Segment:** one agent divides the document into logical parts.

**Budget:** dense sections, such as proofs, get more computing time than an introduction.

**Verify:** separate agents check each part in parallel, with the whole document available for cross-references.

**Synthesise:** a final agent merges the reports, re-checks consistency and grounds claims with external sources, to catch its own hallucinations.

## Who can use it

Everyone gets two free reviews a week, starting with PDFs and plain text; heavier users can plug in a Gemini API key. Google says uploads are never used to train its models and are kept for a year so you can see your results. Beyond papers, it pitches the tool at design documents, slide decks and legal or financial documents.

## Why it matters

Peer review is under real strain, and an automatic first pass for logic and maths errors could help, especially for researchers without senior colleagues to read their drafts. Google is explicit that higher-level judgement stays with humans. That is the right line: a tool like this should make reviewers sharper, not replace them.

## Sources

- [Google Research: Analyzing technical documents with Deep Review](https://research.google/blog/analyzing-technical-documents-with-deep-review/)
- [Google: Research@ 2026, novel capabilities and intelligence](https://blog.google/innovation-and-ai/technology/research/research-capabilities-intelligence-2026/)