This article examines the issue of using authentic speech data in an experimental research paradigm. We report exploratory studies to replicate a seminal investigation of listeners’ abilities to predict sentence completion in constructed read-aloud data. Our initial intention was to see whether the same gating instrument used on authentic talk from interactive settings would produce similar results. In the research ‘journey’ to find natural speech data and match experimental procedures, we came to ask whether the requirements of the laboratory to isolate and decontextualize talk facilitate findings that will truly illuminate interactive talk in natural settings. We suggest that the experimental paradigm struggles to engage with the multi-faceted interpretive tasks which participants engage with in actual talk. Our small-scale studies offer two key conclusions for further work: that a feature of unfolding talk derived under experimental conditions may be strongly predictive, but may only be so under these conditions; that it should not be assumed that the experimental approach can yet do justice to a basic feature of spoken discourse: interactional negotiation over utterance completion.