Summarising User Interviews Without Losing The Quotes

The problem with summarised research is that it launders specificity into consensus. "Users want better onboarding" is not something any user said.

Constraint that fixes it:

"Summarise these 9 interview transcripts.

Rules:

• Every theme must include at least two verbatim quotes with the participant ID

• Count how many participants each theme came from; do not write 'users' for something two people said

• Keep contradictions as contradictions — do not average them into a middle position

• List anything mentioned exactly once separately, under 'signal or noise'

• Do not include recommendations"

The contradictions rule matters most. Two users wanting opposite things is real information about a segment split. Averaged into one bland statement it becomes a feature that satisfies neither of them.

Banning recommendations keeps the summary a summary. I want to read the evidence before someone tells me what to do about it.