Behavioral science is overly focused on a small subset of the global population. This leads to a “generalizability problem” because the degree to which behavioral science theories adequately capture the psychology of populations outside this small subset is unknown. “Big team science” projects, which involve huge teams of collaborators from research labs all over the world that run participants using a common protocol1, have been heralded as a solution to this “generalizability problem” because they provide a way for researchers to recruit participants from many regions rather than just one. Here we evaluate the adequacy of one such big team science project which claimed the “globalizability” of temporal discounting, the phenomenon in which the subjective value of deferred rewards is smaller than immediate rewards2. Although the project recruited 171 researchers from 109 institutions, and 13629 research participants speaking 40 languages across 61 countries, we argue that relying solely on the typical big team methodology created an “illusion of generalizability”, leading authors to overestimate the extent to which research findings can be applied globally. We found that, across low-, middle-, and high-income countries, samples were all similarly young, well-educated, urban, digitally-connected, and the authors were largely concentrated in the Global North. This homogeneity belies the heterogeneity present within each country3,4. We argue that researchers using big team science methods should more closely attend to how their methods include or exclude underrepresented populations, thereby avoiding the “illusion of generalizability”.