audiotranskription Hintergrund

AI Research

Procedures & Resources

LLMs can generate text that attempts to simulate observations, questions, and interpretations. It seems helpful to make use of these. The key factor is how these contributions are integrated into a transparent, material-based analytical process.

On this page, you’ll find our two proposed approaches—Hybrid Interpretation and Serendipity Prompting—each with a methodological foundation, instructions, and a prompt for you to try out. In addition, here’s a guide on how to cite and document the use of AI, and a collection of prompts for different phases of your research project.

Hybrid Interpretation with Dialogically Integrated LLMs

What's this about?

In hybrid interpretation, multiple LLMs work sequentially and in coordination with one another on a selected text passage.

The researchers facilitate this process, review the proposals based on the source material, and decide which interpretations to pursue, modify, or reject.

The entire dialogue is documented; this makes it possible to trace how interpretations arise and are justified.

  • An Overview of the Method [Article]
    The article explains the roles, process, and facilitation of the procedure.
    A simulated interview segment demonstrates how proposed interpretations evolve over several rounds, are tested against the source material, and are refined into a unique interpretation. The introduction is supplemented by screenshots, information about the services used, and a podcast episode.
  • Methodological Framework [Preprint]
    The preprint develops the methodological rationale for the procedure and demonstrates how LLM contributions can be integrated into an interpretive-reconstructive analysis without shifting interpretive authority.
  • Book chapter in English [Eng.]
    This article presents the procedure to an international audience and discusses its methodological prerequisites and practical applications.
  • For Teaching [Book Chapter]
    The teaching exercise (Chapter 6.8, starting on p. 287) demonstrates how students develop their own interpretations, compare them with LLM-generated responses, and justify their decisions based on the material.

Give it a try

  • Copy prompt
    The prompt template outlines the roles, procedure, and documentation for the interpretation session. A version date makes it easier to include this information in your own method documentation.
  • Step-by-Step Instructions (PDF)
    The guide walks you through a complete workflow using browser-based LLMs. It is suitable for practice material or data that does not contain personal information
  • Documented Sample Analysis (DOCX)
    Complete interpretation process with prompts, model contributions, and moderation decisions, as a reference and template for your own documentation

For real interview data, you should clarify in advance under what conditions AI processing is even permitted.

Document in f4

In f4, you can link text passages, codes, memos, and comments to one another.

For each round of interpretation, the prompt, model response, review, and your own conclusion can be documented directly in the material.

This way, the use of AI becomes part of the analysis project rather than an external chat history that is difficult to track.

Serendipity Prompting

What's this about?

Serendipity prompting begins before the actual interpretation takes place.

The LLM identifies a few linguistic or content-related anomalies and poses open-ended questions rather than formulating an interpretation.

These disturbances can increase theoretical sensitivity. Whether they are analytically sound will only become clear as we continue to work with the material.

  • An Overview of the Method [Article]
    This article explains the concept of the LLM as a catalyst, the structure of the prompt architecture, and its limitations.
    The latest version of the prompt is available directly there
  • Idea and Concept [Paper in press]
    The paper develops a theoretical framework for the method and examines its application to different types of models

Give it a try

  • Serendipity Prompt to Copy
    The prompt intentionally limits the output to just a few observations and questions. The results should be viewed as suggestions, not as an analysis.

The same requirements apply to sensitive material as to other AI processes.

Document in f4

The ideas are recorded as memos at the relevant points in the text.

It is important to document whether and how these findings were incorporated or rejected in the subsequent analysis process.

Citing & Documenting AI

What's this about?

LLM outputs are not reliable, independently verifiable sources.

What matters, therefore, is not citation in the traditional sense, but rather transparent documentation of the use: model, prompt, purpose, selection, and verification.

Give it a try

  • Practical Guide
    This guide explains how to describe the use of AI in the methods section, footnotes, and indexes—from transcription to interpretation.

Collection of Prompts

Even though prompts can quickly become outdated and aren’t the be-all and end-all, we’ve collected a few that might inspire you and encourage you to experiment. The collection includes prompt templates for various stages of the research process. They are intended as working tools; they do not replace your own methodological decisions and will almost certainly need to be further adapted and supplemented.