Our serendipity prompting approach (Prof. Dr. Uwe Krähnke, Thorsten Pehl, Dr. Thorsten Dresing) reverses the usual use of AI: Instead of serving as an answer engine to generate plausible syntheses, the LLM is used as a maieutic prompter .
A specific prompt architecture instructs the AI not to confirm patterns in the data material, but to identify contradictions, areas of tension and contrasts. This output acts as a methodological disturbance that irritates established perspectives and inspires researchers to ask new, abductive questions.
The goal is a “Phase 0” of the analysis that counteracts the “proving-the-obvious” problem and strengthens the analytical agency of the researchers instead of substituting it.
Scientific publication
- Status: In preparation (expected to be published in Q3/2025). The full paper with theoretical derivation, methodological development and systematic validation of various open source models has been completed and will be submitted for publication shortly.
- Authors: Prof. Dr. Uwe Krähnke, Thorsten Pehl, Dr. Thorsten Dresing
- Institutions: MSB Medical School Berlin & audiotranskription.de
Prompt to try it out:
The Serendipity prompt developed and validated in the course of the publication is available for download here. This means that everyone can test it directly with their own materials. It has been successfully tested on large commercial LLMs (claude, gemini, chatGPT), opensource LLMs (qwen3-235b) as well as small opensource LLMs (e.g. gemma3-27b). You use this prompt by entering it together with the data material in the chat window of an LLM and sending it together (please note GDPR if applicable).
Contact: info@audiotranskription.de | Status: 23.07.2025
