bdautomated has released a free reference page and dataset tracing 75 widely cited AI agent statistics back to their original sources. This analysis reveals that statistics often disagree because they measure different things, such as intent versus actual use. Surveys counting any adoption or intent report significantly higher figures than those measuring scaled implementation. For instance, while McKinsey found 23 percent of organizations had scaled AI agents, PwC reported 79 percent of U.S. executives believed they were already being adopted. Real-world use in any single department is consistently reported as no more than 10 percent. The analysis also clarifies the MIT Project NANDA statistic, explaining it measures profit-and-loss impact within six months of a pilot, not general AI project failure. Gartner's cancellation forecast is presented as a future projection, not a current count. Every statistic on the page underwent four checks to ensure accuracy and transparency regarding what was measured, who was surveyed, and when. Market-size forecasts were excluded as their underlying reports were not publicly accessible for verification. The dataset is freely available in CSV and JSON formats under a CC BY 4.0 license. bdautomated aims to provide clarity for business owners to understand which statistics are relevant to their specific situations.