# RIT-RAG preprint reports higher accuracy in large document searches

_Published Monday, October 19, 2026 at 4:06 AM EDT · AI, Science · Latest · Tier 2 — Notable_

Researchers report that RIT-RAG, a method for answering questions from document collections, improved accuracy by 6.6-11.4 points over the strongest baseline across three language models on EntQABench. Their new benchmark contains 2.84 million technical-documentation webpages.

The method builds a tree for each document from its table of contents or sitemap. At query time, retrieved passages identify smaller branches across potentially multiple documents; an AI agent then selects nodes to read and reformulates searches when needed. The preprint also reports the highest answer accuracy among tested retrieval baselines on financial, scientific and customer-support benchmarks. These are benchmark results rather than demonstrated production performance.

## Sources

- [cs.AI updates on arXiv.org](https://arxiv.org/abs/2610.11370)

---
Canonical: https://techandbusiness.org/newswire/LG9NgjlkNMGYMTnAVElQLv
Published: 2026-10-19T08:06:40.946Z
Story chronology: 2026-10-19T04:00:00.000Z
Retrieved: 2026-10-19T11:06:12.652Z
Publisher: Tech & Business (techandbusiness.org)