AI-powered Fact-Checking Platform for Local News Verification

Thesis Abstract

This study addresses the pervasive challenge of misinformation in local news ecosystems by developing and evaluating an AI-powered fact-checking platform designed to verify geographically localized reporting in real-time. The problem centers on the time lag, limited coverage, and variable accuracy of local fact-checking due to resource constraints in newsroom operations, coupled with the rise of user-generated content disseminated through community channels. The aim is to design, implement, and assess a scalable platform that integrates machine learning-based claim detection, knowledge-base verification, and narrative-consistency analysis to support journalists, community publishers, and informed citizens. Specific objectives include (1) to develop a modular architecture that combines claim extraction, source credibility scoring, and evidence retrieval from structured knowledge bases and local data repositories; (2) to evaluate the platform’s accuracy, speed, and usability in real-world local news contexts; (3) to examine the platform’s impact on journalists’ verification practices and editorial decision-making; and (4) to explore the ethical and governance implications of automated fact-checking in local news ecosystems. A mixed-methods research design is employed, incorporating quantitative experimentation and qualitative inquiry. The population comprises local newsrooms, community media outlets, and citizen-user groups in two metropolitan regions with diverse linguistic and socio-political profiles. A purposive sample of 12 local outlets and 200 resident participants is selected, with 8 outlets providing access to archival local stories and 4 acting as pilot testing sites for the platform over a six-month period. Data collection instruments include (a) a claims-detection and verification module integrated into the platform, (b) a structured newsroom workflow survey and semi-structured interviews with editors and reporters, (c) a usability and perceived usefulness instrument for citizen users, and (d) an auditable dataset of verified and falsified local news items annotated by professional fact-checkers. Validity and reliability are ensured through triangulation, inter-rater reliability for human annotations (Cohen’s kappa ? 0.75), and pilot testing to refine algorithms prior to full deployment. Analytical methods include logistic regression to model the likelihood of verification success given platform-assisted evidence, time-to-verification analysis using survival analysis techniques, and ANOVA to compare verification performance across outlets and user groups. Thematic analysis will be applied to interview transcripts to identify perceived value, trust dynamics, and workflow barriers, guided by the Technological Frames Theory and the Knowledge-Based View. A conceptual model will be tested that integrates algorithmic credibility, source diversity, and narrative consistency as determinants of verification outcomes. Key expected findings include improvements in verification speed and accuracy when the platform’s evidence retrieval and cross-source corroboration components are activated, with a projected 18–25% reduction in time-to-verification and a 12–15 percentage point increase in correct verifications compared to manual methods. The platform is anticipated to influence newsroom practices by increasing routine verification checks, improving traceability of claims, and expanding source diversity beyond traditional outlets. Among citizen users, it is expected to enhance discernment of local claims and increase trust in local reporting when transparent provenance and evidence links are presented. The study will also identify tensions related to automation biases, data privacy concerns, and potential adversarial manipulation of knowledge bases. The contribution to knowledge lies in providing an empirically grounded, scalable framework for AI-assisted local news verification, integrating theory-driven design with practical newsroom workflows, and offering a model for assessing impact on trust, workflow efficiency, and community information resilience. The main conclusion is that an openly auditable, modular AI fact-checking platform can meaningfully augment local journalism’s verification capacity while preserving journalistic judgment and community trust, provided that human-in-the-loop governance, transparent provenance, and continuous updates to the knowledge base are maintained. Recommendations include (1) adopting a hybrid verification protocol that combines automated evidence gathering with mandatory human adjudication for high-stakes claims; (2) investing in ongoing knowledge-base curation from local sources to reduce reliance on external authority gaps; (3) establishing clear governance policies on data privacy, transparency, and accountability; and (4) scaling the platform to other local ecosystems with context-sensitive customization and multilingual support where necessary.