A comprehensive study examined the political discourse in synagogue sermons in the United States in recent years.
By: Dr. Ghila Amati, Shlomi Bereznik
Methodology
This study draws on a dataset of 4,302 American Jewish sermons, including 2,556 delivered between October 2021 and October 2024 across Reform, Conservative, and Modern Orthodox congregations. Sermons were collected through a combination of automated transcription tools and direct outreach, standardized into a unified format to enable systematic analysis across denominations and time periods.
The analytical framework employed advanced computational discourse analysis techniques, using AI tools such as ChatGPT-4o. Methodologically, the study relied on prompt engineering, few-shot learning, decomposed prompting, and Chain of Thought (CoT) reasoning to structure the AI’s interpretive processes. Sermons were analyzed individually to preserve contextual integrity, and AI-generated outputs were formatted in structured JSON for consistency.
The analysis proceeded in stages: identifying political content; detecting criticism of Israel; classifying sentiment and structure; and assessing specific topics like hostages, ceasefire, and Aliyah. To validate findings, a human-in-the-loop (HITL) approach was adopted, supplemented by standard metrics (accuracy, precision, recall) to evaluate AI performance.