AI-Powered Fact-Checking for Local News Ecosystems

Thesis Abstract

The rapid spread of misinformation through local news channels undermines public trust, democratic accountability, and the social fabric of communities, particularly in the absence of scalable verification mechanisms that can operate at the speed of contemporary digital news ecosystems. This study addresses the persistent gap between access to diverse local information and the reliability of that information by developing and evaluating an AI-powered fact-checking framework tailored to local news contexts. The aim is to design, implement, and assess a scalable system that integrates automated claim verification, source credibility assessment, and audience-facing transparency features to bolster the integrity of local journalism. Specific objectives include (1) profiling local news environments to identify prevalent misinformation typologies and information flows, (2) developing an end-to-end AI-driven fact-checking pipeline that combines natural language processing, citation network analysis, and multimodal evidence retrieval, (3) evaluating user trust, perceived credibility, and cognitive load associated with automated fact-checks among local news readers, and (4) examining the impact of a transparency dashboard on newsroom editorial practices and audience engagement. The study is guided by the Communication Theory of Verifiability and the Credibility Theory, complemented by the Technological Frames approach to understand how journalists and audiences adopt AI-based verification tools. A mixed-methods research design is employed. The quantitative strand uses a stratified sample of 60 local news outlets and 1,200 readers across three metropolitan regions, with a 12-month longitudinal component to capture implementation and reception dynamics. The qualitative strand comprises 30 in-depth interviews with editors, reporters, and fact-checking personnel, and 12 focus groups with local news consumers. Data collection instruments include (i) a calibrated fact-checking toolkit incorporating an automated claim classifier, a source credibility scorer, and an evidence retrieval module; (ii) newspaper and online article corpora containing 5,000 local news items annotated for veracity by human fact-checkers; (iii) survey instruments measuring trust, perceived accuracy, and cognitive effort; and (iv) interview and focus group protocols anchored in semi-structured questions. Validity and reliability are established through triangulation, inter-rater reliability checks for human annotations (Cohen’s kappa > 0.75), pilot testing (n=100 readers), and cross-validation of the AI model on held-out data. Analytical techniques include descriptive statistics and regression analyses to identify predictors of user trust and engagement; event-study methodology to assess changes in newsroom practices following tool deployment; thematic analysis of qualitative transcripts to extract themes related to adoption barriers and governance, and network analysis to map information flows and verification patterns within local ecosystems. The analytical model Specification includes a hierarchical linear model to account for nested data (readers within regions within outlets) and a structural equation model to test the relationships among tool usability, perceived credibility, and user engagement. Expected findings indicate that an integrated AI fact-checking pipeline reduces the spread of locally contextual misinformation by up to 28% in flagged content, enhances reader trust scores by 15–20 percentage points, and prompts measurable shifts in editorial processes toward proactive verification, without increasing production time by more than 2–3 minutes per item. The study anticipates that transparency dashboards, when aligned with newsroom workflows, improve audience perception of accountability and augment engagement metrics (time on article, share rates). It is anticipated that limitations will include regional variability in data availability and potential biases in source credibility assessments, mitigated by continuous model retraining and human-in-the-loop verification. The contribution to knowledge lies in (i) advancing a practically deployable, AI-enabled verification framework tailored to local news ecosystems; (ii) empirically linking verification technology with newsroom governance and audience trust dynamics; and (iii) offering a theoretical synthesis of verifiability, credibility, and technological frames in the context of automated fact-checking. Recommendations include scaling the framework to multilingual localities, establishing independent governance for AI verifications, and developing policy guidelines to ensure transparency, accountability, and sustainability of AI-powered fact-checking in local journalism. A roadmap for future work emphasizes adaptive learning from user feedback, expansion to audiovisual misinformation, and integration with platform-level content moderation standards.