How conversational AI shapes the readability of 10-K reports

Purpose We investigate whether the adoption of conversational artificial intelligence (CAI) translates corporate complexity into clearer financial narratives. CAI, representing AI-driven innovation, can bolster operational efficiency and performance, thereby leading to better financial transparency and readability of financial statement information. Design/methodology/approach We construct a firm-level CAI measure from CAI-related disclosures in 10-K filings using Sent-latent Dirichlet allocation (LDA)-variational expectation maximization (VEM) topic modelling. Our measure captures CAI applications related to service and operations, and decision-making and governance. We also use patent-based measures based on natural language processing and speech-recognition technologies as alternative proxies in robustness tests. Findings We find that CAI adoption is associated with lower Bog Index scores, indicating more readable 10-K reports. Mediation tests show that financial performance and operational efficiency partially mediate this relationship. The results remain robust after addressing endogeneity concerns and using alternative specifications. Originality/value Using unsupervised machine learning topic modelling, we observe that CAI is predominantly employed in corporate operations, notably in service-related activities, decision-making processes and corporate governance.