[Paper Review] CiteAudit: You Cited It, But Did You Read It? A Benchmark for Verifying Scientific References in the LLM Era

Zhengqing Yuan, Kaiwen Shi+4 more|arXiv (Cornell University)|Feb 26, 2026

Scientific Computing and Data Management0 citations

TL;DR

CiteAudit introduces a multi-agent framework and a large-scale benchmark to verify the faithfulness and evidence alignment of cited references in scientific writing, addressing hallucinated citations in the LLM era. It shows improved detection accuracy and interpretability over baselines and provides standardized evaluation protocols.

ABSTRACT

Scientific research relies on accurate citation for attribution and integrity, yet large language models (LLMs) introduce a new risk: fabricated references that appear plausible but correspond to no real publications. Such hallucinated citations have already been observed in submissions and accepted papers at major machine learning venues, exposing vulnerabilities in peer review. Meanwhile, rapidly growing reference lists make manual verification impractical, and existing automated tools remain fragile to noisy and heterogeneous citation formats and lack standardized evaluation. We present the first comprehensive benchmark and detection framework for hallucinated citations in scientific writing. Our multi-agent verification pipeline decomposes citation checking into claim extraction, evidence retrieval, passage matching, reasoning, and calibrated judgment to assess whether a cited source truly supports its claim. We construct a large-scale human-validated dataset across domains and define unified metrics for citation faithfulness and evidence alignment. Experiments with state-of-the-art LLMs reveal substantial citation errors and show that our framework significantly outperforms prior methods in both accuracy and interpretability. This work provides the first scalable infrastructure for auditing citations in the LLM era and practical tools to improve the trustworthiness of scientific references.

Motivation & Objective

• Motivate the need to combat hallucinated citations in scholarly writing due to LLMs.
• Propose a scalable, multi-agent verification framework to assess citation faithfulness and evidence alignment.
• Create a large, human-validated benchmark spanning diverse domains and citation types.
• Provide unified evaluation protocols and metrics for citation verification.
• Demonstrate improved accuracy and interpretability over baselines in experiments with state-of-the-art LLMs.

Proposed method

• Design a five-agent pipeline: Claim Extractor, Retriever, Evidence Matcher, Reasoner, and Judge, coordinated by a planning controller.
• Develop a large-scale dataset combining real-world citation errors and systematically generated hallucinated ones with human validation.
• Formalize verification as a multi-stage evidence-consistency task with a strict verification criterion requiring exact metadata matches.
• Implement web-search and scholarly databases as external knowledge sources via agents to ground evidence.
• Evaluate models on both a generated benchmark and a real-world test set using standard metrics for citation faithfulness and verdict consistency.

Experimental results

Research questions

• RQ1 Can a multi-agent framework reliably detect hallucinated citations in scientific manuscripts?
• RQ2 How does evidence retrieval and reasoning affect faithfulness judgments across diverse citation types?
• RQ3 What is the impact of incorporating authoritative scholar verification on recall and precision?
• RQ4 How well do generated benchmarks reflect real-world citation error patterns in practice?

Key findings

• The benchmark combines real-world and synthesized hallucinated citations, showing realistic error patterns similar to those observed in practice.
• Multi-agent verification significantly improves accuracy and interpretability over single-model baselines.
• The Scholar Agent as a final verification stage reduces resilient hallucinations that pass web-based checks alone.
• On real-world data, the proposed framework achieves the highest accuracy, precision, recall, and F1 among evaluated methods.
• The approach incurs lower cost and latency than many proprietary LLM-based solutions by limiting heavy reasoning to planning and final judgment stages.