It is widely hoped that statistical models can improve decision-making related to medical treatments. Because of the cost and scarcity of medical outcomes data, this hope is typically based on investigators observing a model’s success in one or two datasets or clinical contexts. We scrutinized this optimism by examining how well a machine learning model performed across several independent clinical trials of antipsychotic medication for schizophrenia. Models predicted patient outcomes with high accuracy within the trial in which the model was developed but performed no better than chance when applied out-of-sample. Pooling data across trials to predict outcomes in the trial left out did not improve predictions. These results suggest that models predicting treatment outcomes in schizophrenia are highly context-dependent and may have limited generalizability. , Editor’s summary A central promise of artificial intelligence (AI) in healthcare is that large datasets can be mined to predict and identify the best course of care for future patients. Unfortunately, we do not know how these models would perform on new patients because they are rarely tested prospectively on truly independent patient samples. Chekroud et al . showed that machine learning models routinely achieve perfect performance in one dataset even when that dataset is a large international multisite clinical trial (see the Perspective by Petzschner). However, when that exact model was tested in truly independent clinical trials, performance fell to chance levels. Even when building what should be a more robust model by aggregating across a group of similar multisite trials, subsequent predictive performance remained poor. —Peter Stern , Clinical prediction models that work in one trial do not work in future trials of the same condition and same treatments.