Abstract
News media's use of artificial intelligence (AI) has prompted regulatory activity (e.g., NY FAIR News Act), which typically recommends editorial oversight and transparency regarding AI use as accountability measures. But do these measures together help outlets win readers? Since journalists typically lack full agency over content despite driving revenue, we study this question using a principal-agent framework that links production (editor-journalist) and market (reader-outlet) models through content quality, with agents privately differing in AI-use intensity and principals enforcing accountability through oversight and transparency. We analyze the production and market models individually and jointly. Individually, we find that while costless oversight is always recommended, costly oversight can limit the oversight rents that heavy AI users extract. Further, transparency demand is not unconditional but depends on the content's alignment with readers' beliefs. Indeed, the so-called "transparency discount" is not guaranteed, and readership can increase with transparency when readers value news more than belief alignment. Yet, joint analysis reveals a transparency inversion paradox: an outlet can gain readership by disclosing AI use that degrades content, but lose readership by disclosing use that improves content. Thus, our study demonstrates regulatory weaknesses that can help advise recent legislative efforts on AI use in media.