In our first blog post in this series, we documented that U.S. workers are more likely to use artificial intelligence (AI) on the job than workers in Europe. In our second post, we showed that differences in workforce composition and firm management practices go a long way toward explaining that gap. But alongside the worker surveys, we drew on firm-level data, and those data address a puzzle of their own—one that turns out to be less about economics and more about how surveys are designed. For the past two years, a number of researchers have noted a striking disconnect in the U.S. data: Worker surveys suggest that somewhere around 35% to 40% of workers use AI on the job, while the main U.S. firm survey put AI adoption among businesses at just 5% to 7%. That is a very large gap. Are workers using AI without their employers knowing? Are firms in denial about how widespread the technology has become? Or is something else going on? We think the answer is mostly the third option: measurement.