Since 2023 most of my development work has been about getting AI to do parts of qualitative research that used to take weeks of hand coding, without giving up the ability to check every claim against the transcript it came from.
Causal Map uses AI to propose causal links from interview text. The evaluator keeps the coding frame, edits the links, and every link on the map stays attached to the sentence it came from. We tested how well an untrained model does this against expert human coders: AI-assisted causal mapping.
QualiaInterviews is an AI interviewer. You write the interview guide. It then interviews each respondent, following up where a human interviewer would, and gives you the transcripts. It is for the cases where a survey is too thin and fifty face-to-face interviews are too expensive. There is background on what it does and a piece on the potential of AI interviewing. Researchers who already have an account sign in at manage.qualiainterviews.com.
The two fit together as one pipeline, from AI interview to coded causal map: A workflow for collecting and understanding stories at scale, in Evaluation 31(3), 394 to 411 (2025), summarised here.
On doing it rigorously
This is the part I care about most. I argue it out at length in the AI in qualitative social science chapter of the Garden.
- Just add rigour: three do’s and don’ts. Break a vague task into small ones somebody else could check.
- Trust the algorithm, not the AI.
- Put down that thesaurus, an open call to qualitative researchers who object to AI on grounds that would also rule out a thesaurus.
Most weeks I take one of these arguments further in my LinkedIn newsletter, Causal mapping, evaluation, AI.