Corporate attention to generative AI and earnings management: evidence from China

This study examines how firms’ strategic attention to generative artificial intelligence (GenAI) is associated with three forms of earnings management. Using 25,208 firm-year observations for Chinese A-share firms from 2011 to 2023, we measure firm-level GenAI-related strategic attention with a term frequency-inverse document frequency (TF-IDF) index constructed from Management Discussion and Analysis (MD&A) disclosures. Empirical results show that higher GenAI-related attention is associated with lower real earnings management (REM), no significant change in accrual-based earnings management (AEM), and higher textual earnings management (TEM), measured by abnormal optimistic tone. The results remain robust to alternative measures, the 2015–2023 subsample, high-dimensional fixed effects, and an industry-peer instrumental-variable specification. Mediation analysis shows that the administrative expense ratio partially mediates the positive association between GenAI-related attention and TEM: greater GenAI-related attention is associated with a lower administrative expense ratio, which in turn is associated with higher TEM. Heterogeneity analysis shows that the negative REM relation is concentrated among less diversified, domestic, and non-outsourcing firms. The positive TEM relation appears in both diversification groups but only in domestic and non-outsourcing firms. Given the significant positive association between GenAI-related attention and TEM, we further examine whether TEM has capital-market consequences. The results show that TEM predicts higher stock price crash risk in the following year. Overall, the findings are consistent with a shift in managerial discretion from operational actions toward narrative disclosures rather than a uniform improvement in reporting quality.