Artificial intelligence (AI) is emerging as a general purpose technology (GPT) with the potential to transform industries. As a result, the potential benefits of AI give firms strong reasons to adopt it – but also provides opportunities to exaggerate their investments in it. Unlike previous GPTs such as the steam engine or electricity, which required investments in physical infrastructure, AI depends heavily on intangible investments, including human capital, organizational restructuring, and firm-specific data capabilities (Brynjolfsson, Rock, and Syverson, 2019; Bresnahan et al., 1996). These investments are difficult to quantify and often underreported in traditional financial filings, creating uncertainty about the actual extent of firms’ AI investments.

Despite these challenges, investor interest in AI has grown, and firms that make these investments in AI often benefit from premium valuations (Babina et al., 2024). This has motivated firms to disclose their AI capabilities in 10-K filings, earnings statements, and conference calls. In a new paper, we show that, while some firms genuinely invest in AI-driven innovation, others may exaggerate or misrepresent their AI capabilities to attract investors – a practice referred to as “AI washing.” Former SEC Chair Gary Gensler warned firms about overstating their use of AI, and the commission has launched enforcement actions against suspected cases of AI washing (e.g., SEC, 2019).[1]

Specifically, in the paper, we examine the determinants and informativeness of AI disclosures, focusing on whether public AI disclosure matches the level of investment in employees with AI skills or roles. Our study is guided by four central questions: (1) What types of firms are more likely to disclose AI activities, (2) do these disclosures provide informative insights into firms’ use of AI and their future efficiency, innovation, and dividend payout policy, (3) how do firms that disclose extensive AI activities but don’t invest much in employees who work with AI – —which we label as suspected AI “washers” – differ from non-washers regarding firm characteristics and informativeness, and (4) what capital market outcomes are associated with AI disclosure, and do they differ for firms suspected of AI washing?

To measure AI disclosure and washing, we fine-tune a natural language processing model (e.g., FinBERT), to examine conference calls, 10-K filings, and earnings announcements from 2016 to 2023. We complement this with firms’ AI-related employment data from Revelio as a proxy for complementary AI investments (similar to Babina et al. 2024). Following the greenwashing and diversity literature (Baker et al., 2024), we classify firms as suspected AI washers if they fall into the highest tercile of AI disclosure but the lowest tercile of AI employment.

We offer four main findings. First, AI disclosures are more common among firms in AI-intensive industries, those with high innovation, and those facing greater investor scrutiny. Second, AI disclosures positively correlate with future operational efficiency and AI patent filings but negatively correlate with dividend payouts, suggesting firms reinvest AI-driven gains rather than distributing cash. Third, suspected AI washers – those disclosing AI without corresponding AI employment – don’t have these outcomes and tend to be smaller, less innovative, and in non-AI-intensive industries. Finally, firms making AI investments outperform AI washers in long-term abnormal returns, highlighting the importance of in-house employees involved with AI tasks.

Several additional analyses support the robustness of our findings. Our results remain consistent when we use alternative AI washer definitions and control for firms’ greenwashing. The findings also hold for outcomes two and three years ahead, suggesting that timing differences between investment and results do not explain our observations. Furthermore, extensive AI disclosure combined with substantial investment in workers involved with AI predicts improved efficiency and more AI patents, while the absence of both elements predicts reduced efficiency and fewer AI patents, suggesting that meaningful AI implementation requires both significant disclosure and matching labor investment.

Our study makes several contributions to the literature. First, we extend AI capital markets research by examining how firms communicate AI adoption through disclosures. Our findings show that AI disclosure predicts future firm performance and stock prices, but only when backed by credible AI-related labor investments. Second, we provide some of the first empirical evidence on AI disclosure and AI washing, addressing growing regulatory concerns about washing from the SEC. Our study shows how market pressures influence disclosure around emerging technologies, particularly intangible AI investments. Our findings also contribute to the broader literature on misleading corporate disclosures, drawing parallels to greenwashing and diversity washing. However, unlike ESG disclosures – where investor reactions often depend on psychological framing or social impact considerations – AI disclosures are more directly linked to financial outcomes through their impact on firm efficiency and product capabilities. We find that AI washing is not simply a subset of ESG washing, as firms suspected of ESG misrepresentation do not strongly correlate with those engaging in AI washing, suggesting that AI-related misstatements are driven by distinct incentives and market dynamics.

Third, we contribute to the labor and accounting literature by examining how firm-level AI-related employment intersects with corporate disclosure strategies. Unlike prior waves of automation, AI primarily affects white-collar jobs, complementing high-skilled workers while substituting lower-skill roles. Our key finding is that AI-capable employees are integral in extracting value from AI, providing insights into how firms navigate AI adoption and how markets interpret these signals.

Our findings highlight that AI disclosures provide valuable market signals, but only when paired with real investments in AI-related workers. As AI adoption accelerates, distinguishing between genuine AI integration and strategic misrepresentation will be critical for investors, regulators, and policymakers assessing firm value and the broader economic impact of AI.

ENDNOTE

[1] https://www.bleepingcomputer.com/news/technology/investment-advisers-pay-400k-to-settle-ai-washing-charges/?trk=feed_main-feed-card_feed-article-content

REFERENCES

Babina, T., A. Fedyk, A. He, and J. Hodson. 2024. Artificial intelligence, firm growth, and product innovation. Journal of Financial Economics 151.

Baker, A.C., Larcker, D.F., McClure, C.G., Saraph, D. and Watts, E.M., 2024. Diversity washing. Journal of Accounting Research, 62(5), pp.1661-1709.

Bresnahan, T., S. Greenstein, D. Brownstone, and K. Flamm. 1996. Technical progress and co- invention in computing and in the uses of computers. Brookings Papers on Economic Activity: Microeconomics: 1-83.

Brynjolfsson, E., D. Rock, and C. Syverson. 2019. Artificial intelligence and the modern productivity paradox: A clash of expectations and statistics. The Economics of Artificial Intelligence. University of Chicago Press: 23-60.

SEC. 2019. Comment letter regarding Progenics Pharmaceuticals, Inc.’s PREC14A filed on September 25, 2019. Securities and Exchange Commission.

This post comes to us from John M. Barrios at Yale University School of Management, John L. Campbell at the University of Georgia’s J.M. Tull School of Accounting, Ryan G. Johnson at Indiana University’s Kelley School of Business, and Christine Liu at Bentley University. It is based on their recent article, “Artificially Intelligent or Artificially Inflated? Determinants and Informativeness of Corporate AI Disclosures,” available here.