Survey Maps How AI Language Models Are Reshaping Automated Fact-Checking

The challenge of verifying facts at scale has become urgent as false and misleading content spreads rapidly across digital platforms, outpacing the ability of human fact-checkers to keep pace. A new systematic survey examines how generative large language models (LLMs) are being deployed to automate and accelerate fact-checking workflows.

The Evolution of Fact-Checking with LLMs#

Generative LLMs have expanded the toolkit for combating misinformation beyond simple accuracy prediction. These models now serve as critical enablers across the entire fact-checking pipeline—they help generate training data, assist in verifying claims, contribute to evaluating system performance, and support human fact-checkers in their work. The shift represents a meaningful evolution in how researchers approach the problem of identifying and countering false information.

Research Scope and Findings#

The survey analyzed 199 academic papers submitted to arXiv between 2024 and 2026, creating a unifying framework for understanding how generative LLMs integrate into fact-checking systems. The researchers documented patterns across multiple dimensions:

• Task applications: how models perform claim verification, generate new data, and support evaluation
• Dataset strategies: techniques for constructing and expanding training datasets
• Formulations and methods: different approaches to problem framing
• Evaluation practices: how researchers measure system performance

Global Challenges and Research Gaps#

A particular focus of the survey is the state of fact-checking in multilingual and low-resource settings. The findings reveal that while LLMs show promise for addressing misinformation globally, current research has significant blind spots—many systems are heavily concentrated on high-resource languages and well-funded regions, leaving gaps in coverage and accessibility for underrepresented communities.

Looking Forward#

The survey identifies this period as a critical juncture for developing fact-checking systems that are more reliable, transparent, and inclusive. As generative LLMs continue to evolve, researchers face ongoing challenges in ensuring these powerful tools strengthen rather than undermine public trust in information.