Affinity mapper
What it does
Affinity mapping takes a pile of individual observations and groups them by relationship until patterns become visible. The defining property is that it's bottom-up: you don't start with categories, you start with evidence and let the categories emerge. That's what separates it from sorting notes into buckets you'd already decided on. It's also called an affinity diagram, or the KJ method after Jiro Kawakita, who formalised it in the 1960s.
This skill runs that process over transcripts, survey responses or notes you paste in. It writes one observation per note, clusters what relates, names the clusters only once they've stopped moving, and then does the part most teams skip. It breaks the fastest-forming groups apart and rebuilds them. Groups that survive that are real. Groups that don't were your team's existing assumptions wearing a theme name.
What you get back
- Named themes, each with the observations that produced it and the participant each one came from.
- A confidence note per theme: how many people it rests on, and whether it held up when the cluster was rebuilt.
- The orphans: observations that wouldn't cluster, kept rather than dropped. These are often the interesting ones.
- A markdown board you can paste into your repo, or push straight into a Great Question insight.
How to use it
Example Prompts:
"Affinity map the twelve onboarding interviews in this folder."
"Cluster these survey open-ends and tell me which groups are weak."
"Re-run the mapping, but split the 'pricing confusion' theme apart first."
What it won't do
It won't decide what the themes mean for your product. That judgement stays with you, and the skill is built to hand it back rather than guess. It also won't invent evidence to round out a thin cluster: if a theme rests on two people, it says so instead of writing around it.
Questions
What's the difference between affinity mapping and an affinity diagram?
Nothing. Affinity diagram is the term used in service design and quality management; affinity mapping is what UX calls it. Same technique, same origin.
Do I need a Great Question account?
No. The skill works on files you give Claude. If you do have an account, it can pull sessions and transcripts from your repo directly and write the board back as an insight.
How many transcripts can it handle at once?
Up to about twenty in one pass. Past that it maps in batches and merges, which is slower but doesn't lose observations.
