Reception of Human, AI-Generated and Hybrid Sermons: Survey Responses of Polish Priests, Religious and Catechists and Responses of Ten Language Models

This dataset comes from a study of how professional listeners receive sermons produced in different ways. Four sermons on the parable of the sower (Matthew 13:1-23), between 283 and 399 words long, differed only in how they were produced: a human text, an AI-generated text corrected by a human, a text generated entirely by AI, and a human text corrected by AI. In a Google Forms survey (1-9 September 2026), 80 lay catechists, priests and consecrated persons from six Polish dioceses read the sermons without knowing their origin. They rated each sermon on five statements (1-5), answered whether artificial intelligence had been involved in producing it and explained why, chose the best sermon, and said to what extent a preacher may use AI. The same questionnaire was given to ten language models, three runs each. Their answers are in a separate file and should not be merged with the human responses. The dataset contains the anonymised human responses, the model responses, the full texts of Sermons 2, 3 and 4 as shown to respondents, the AI-generated version of Sermon 2 before human correction, and the questionnaire. Sermon 1 is available only as a link to its original publication. All data are in Polish. The README (in English) explains the files, the columns, the origin of each sermon and the known features of the data.