ABSTRACT: Existing experimental evidence on the contact hypothesis has mainly used long and unstructured interventions, with implications for the replicability and scalability of existing contact protocols. We here test the effect of a brief contact, using a structured protocol that can be implemented in a wide range of situations at a reasonable cost. We also evaluate the lasting effects one month after the intervention. Contact is only found to be effective at increasing trust toward the specific individuals met, and only in the short-run. Generic Machine Learning techniques enable us to identify characteristics of the most and least affected groups.