A new study from Carnegie Mellon University and a company called Larridin set out to answer a question that most executives have been guessing at for two years: is any of this AI spending actually paying off in revenue.
The researchers looked at more than 500 large public companies across 12 industries, using financial filings, tens of thousands of job postings, and market data. To make sure the results were not just being driven by chipmakers riding the AI boom, they ran the numbers twice, once excluding Nvidia, Broadcom, AMD, Micron, and Intel, and once with them included. The conclusion held both times.
The key finding is not that using AI helps revenue. It is that talking about AI specifically helps revenue, at least as a signal. Companies were scored from one to five based on how concrete their public statements about AI were. Vague lines like "we are integrating AI across our operations" scored low. Reporting exactly how many employees use a specific AI tool, or what a pilot program actually produced, scored high. Companies in the top tier of specificity grew revenue roughly 8 percent faster than the vaguest talkers.
Visa is the standout case. It disclosed that close to 26,000 employees had started using AI-powered chat tools and described concrete pilots involving AI agents handling transactions. Its revenue rose 17 percent that year. Conagra Brands sits at the other end. It talked about AI and data upgrades in general terms without hard numbers, and its revenue fell 2 percent over the same period.
There is a real catch here, and it matters more than the headline. None of this specificity showed up in operating margins or stock returns. Companies that were precise about AI grew sales faster, but their profitability did not improve any more than their vaguer peers, and their stock did not outperform either. A separate study from a startup called Blue Bridge Group AI, released in June, found the opposite conclusion on stock performance, and Larridin's team is upfront that the two studies used different methods and are not directly comparable.
There is a wrinkle worth watching closely. Regulators, including the SEC, have been paying closer attention to companies that oversell AI capabilities in public filings, a practice now commonly called AI washing. The specific, quantified disclosures that this study rewards with a revenue signal are also exactly what regulators want to see instead of hype.
For any business leader thinking about how to describe AI work internally or externally, the message is straightforward. Generic claims about "embracing AI" do not move the numbers and increasingly carry legal risk if they turn out to be exaggerated. What seems to matter, both to investors and now apparently to revenue itself, is naming the tool, naming the use case, and naming the result. If you cannot say how many people use it and what it produced, you probably have not adopted it as deeply as you think.
The absence of a margin or stock link is also a useful reality check. AI adoption right now looks more like a growth story than a cost-cutting or profit story for most companies, at least based on what shows up in public filings. Anyone expecting AI to quietly fatten margins without real operational change should treat this study as a caution, not a promise.