Predictive analytics and “big data” are emerging as important new tools for diagnosing and treating patients. But as data collection becomes more pervasive, and as machine learning and analytical methods become more sophisticated, the companies that traffic in health-related big data will face competitive pressures to make more aggressive claims regarding what their programs can predict. Already, patients, practitioners, and payors are inundated with claims that software programs, “apps,” and other forms of predictive analytics can help solve some of the health care system’s most pressing problems. This article considers the evidence and substantiation that we should require of these claims, focusing on “health” claims, or claims to diagnose, treat, or manage diseases or other medical conditions. The problem is that three very different paradigms might apply, depending on whether we cast predictive analytics as akin to medical products, medical practice, or merely as medical information. Because big data methods are so opaque, its claims may be uniquely difficult to substantiate, requiring a new paradigm. This article offers a new framework that considers intended users and appropriate evidentiary baselines.