Management Summary Research Study “Do privacy policies build trust in recommender systems?”

Recommender systems — that are algorithms that suggest products, movies, or news to online users — have become an essential component of e-commerce. Their effectiveness depends on users trusting the system and being willing to share personal data with it. Among other measures, companies typically rely on privacy policies to build trust. In two experimental studies, published in Behaviour & Information Technology and Scientific Reports, Marc Schewe and Štěpán Bahník from the Faculty of Business Administration tested whether privacy policies actually fulfill this role and provide reassurance. The results consistently show that they do not. Presenting a privacy policy never increased users’ trust in a recommender system. On the contrary, a long privacy policy reduced trust compared to a short one or no policy at all, and showing any privacy policy made users less willing to share their personal information with the system. Rather than reassuring users, privacy policies might draw attention to privacy risks.

A follow-up study examined whether privacy policies could be made more effective by framing them to explicitly emphasize either the system’s ability to provide accurate and useful recommendations (competence), its good intentions toward users (benevolence), or its honest handling of data (integrity) — the three core components of trust within the Web Trust Model, which was used as the theoretical framework for the research. None of the three framings increased users’ trust, willingness to follow the system’s recommendations, or willingness to share information. For managers, the practical implications are clear: privacy policies are a legal necessity, but they should not be treated as a marketing instrument for building consumer trust. Companies are better advised to request only the data that is strictly necessary, implement strong privacy protections by default, and keep privacy policies concise, with more detailed versions available on request. Trust in personalized services is more likely to be earned through nonaggressive data practices and a repeatedly good user experience with the system than through the content or wording of a policy that most users do not read.

 

Schewe, M., Bahník, Š. (2026). Privacy policy framing and trust in recommender systems: An experimental study. Scientific Reports. https://doi.org/10.1038/s41598-026-66241-6

Schewe, M., & Bahník, Š. (2026). The effects of privacy policy presentation and length on trust in recommender systems: an online experiment. Behaviour & Information Technology, 1-22. https://doi.org/10.1080/0144929X.2026.2686167