Making Genetic Testing Privacy Policies Easier to Read! And Understand? Leveraging LLMs to Improve Readability through Summarization

Privacy policies in direct-to-consumer genetic testing are often written in complex language, too complex for consumers to read and understand. This study presents a multi-stage summarization pipeline that combines feature extraction and multiple language models to generate readable summaries of such policies. We evaluate the summaries in a between-subject user study with 260 participants, across constructs of cognitive load, readability, trust, and understanding. Although conventional metrics indicated insufficient text quality of the DeepSeek and our model, user evaluations show that summaries can reduce cognitive load and improve perceived readability and trust. However, understanding levels remain consistent across all groups comparing our approach, DeepSeek and the full-text. This approach offers a potentially effective solution to the challenges associated with privacy communication in sensitive domains such as genetic testing, while also enhancing consumer perception and effort.