A Comparative Discourse Analysis of AI-Generated vs. Human-Produced News Reports: Evaluating Objectivity, Framing, and Ideology

Abstract

As large language models enter newsrooms, this study compares AI‑generated and human‑produced news on linguistic construction, framing, and objectivity. We analyze N = 360 English‑language reports (AI = 180; Human = 180) across politics, economy, science/technology, and health in the US, UK, and Pakistan (Jan–Jun 2025), combining CDA/SFL‑based coding with computational indicators (lexical diversity/MTLD, syntactic complexity/MLC, readability, subjectivity/sentiment; quotation density, source diversity, reporting verbs). Tests include Welch’s t with Benjamini–Hochberg correction, mixed‑effects models (Origin fixed; Outlet random; Beat/Country controls), binomial GLMs for frames, an Objectivity Index, and an exploratory classifier. AI texts use more modality/hedging and slightly more passive voice but lower lexical diversity and syntactic complexity. The largest gaps are in sourcing (fewer quotes, sources, reporting verbs), yielding a lower Objectivity Index for AI. Human‑Interest framing is more common in human stories, while AI tilts toward Economic Consequences. A classifier distinguishes origin with AUC ≈ 0.96. AI and human news form distinct discourse profiles; pairing AI drafting with sourcing protocols, disclosure, and frame‑aware editing is recommended to meet journalistic standards.