How quickly did GenAI change corporate disclosure, and does the market care?
Ning Jia, Ningzhong Li, Guang Ma, and Da Xu study managerial disclosure about generative AI (GenAI) in their paper « Corporate Responses to Generative AI: Early Evidence from Conference Calls ».
They analyse 6,163 Capital IQ call transcripts from 2,015 firms to determine what drives post-ChatGPT changes in GAI narratives, and whether those narratives inform capital markets.
Their main conclusions include:
- The share of conference calls containing any GenAI discussion rises sharply after ChatGPT’s release, from 0.34% in 2022 Q3 to 9.25% soon after, showing the fast diffusion of GenAI into firms’ communications.
- Five company characteristics are identified as having strong positive relationships with GenAI discussions intensity: innovation intensity, cybersecurity threats, product differentiation, labour exposure, and customer operations.
- Firms in industries facing higher cybersecurity threats or customer operations activities shift towards more negative GenAI talks and are less likely to frame them around concrete action initiatives.
- Firms in industries with greater labour exposure to AI increase initiative-linked GenAI discussion more, consistent with managers treating GenAI as an operational lever where tasks are more automatable.
- The GenAI discussion variation is positively linked to the magnitude of three-day abnormal returns around calls and to abnormal trading volume, both significant at the 1% level.
- Abnormal returns also predicts later increases in GenAI talks, but the effect is absorbed when the five company characteristics variables are included, suggesting they already channels investor beliefs.
This research suggests markets react when GenAI discussion intensity shifts, and separating use cases, risk controls, and implementation timing may help investors translate narratives into cash-flow and risk assumptions.
Opportunity narratives should be separated from governance and operational risk ones, especially in customer-facing and cyber-exposed models.
As limitations, labour exposure and customer operations are proxied at the industry level due to data limitations, the analysis only covers the first months after ChatGPT’s launch, and it infers investor learning from price and volume rather than direct surveys.