In tuberculosis-endemic settings, patients are often treated empirically, meaning that they are placed on treatment based on clinical symptoms or tests that do not provide a microbiological diagnosis (eg, chest radiography). New tests for tuberculosis, such as the Xpert MTB/RIF assay (Cepheid, Sunnyvale, CA, USA), are being implemented at substantial cost. To inform policy and rationally drive implementation, data are needed for how these tests affect morbidity, mortality, transmission, and population-level tuberculosis burden. If people diagnosed by use of new diagnostics would have received empirical treatment a few days later anyway, then the incremental benefit might be small. Will new diagnostics substantially improve outcomes and disease burden, or simply displace empirical treatment? Will the extent and accuracy of empirical treatment change with the introduction of a new test? In this Personal View, we review emerging data for how empirical treatment is frequently same-day, and might still be the predominant form of treatment in high-burden settings, even after Xpert implementation; and how Xpert might displace so-called true-positive, rather than false-positive, empirical treatment. We suggest types of studies needed to accurately assess the effect of new tuberculosis tests and the role of empirical treatment in real-world settings. Until such questions can be addressed, and empirical treatment is appropriately characterised, we postulate that the estimated population-level effect of new tests such as Xpert might be substantially overestimated.