Research & writing · Claude Code

Code interview transcripts into themes with checked quotes

Idea + starter prompt, not run by us yet

For: Researchers and product people sitting on a pile of interview transcripts

Each interview is saved as its own file in one folder, named the same way, next to a CLAUDE.md that lists what the study asks, the codebook, plus a rule that quotes must be exact and name their file. Claude tags them a few at a time or with parallel subagents and writes theme files and evidence tables, then checks each quote in its source. In one check, 7.7% of quotes labelled as exact were not in the source files.

Steps

1. Remove names and personal details from transcripts, since content goes to the model provider.
2. Give every interview its own file and name them in the same pattern.
3. Fix your codebook before starting parallel agents, or theme lists will not match.
4. Have Claude code in small batches and write themes and evidence tables.
5. Always run a pass that checks every quote against its source file.

Starter prompt

The folder [folder] holds [number] interview transcripts with names removed. My research questions are [questions] and my codebook is in [codebook file]. Code the transcripts in batches of [number]. Write a themes file and an evidence table. Only quote exact text and give the file name for each quote. When done, search each quote in its file and list any you cannot find. Do not edit the transcripts.

Original post: ccforeveryone.com · found via agentusecases.ai. Written up in our own words. We haven't run it, so test it on made-up data first.