Back to library

Qualitative Research Synthesiser

The failure mode in AI-assisted research synthesis is confidence without accuracy: the AI produces a polished list of themes that has lost the friction, contradiction, and uncertainty that made the research valuable. This skill keeps AI in the role of categoriser and retriever, and keeps you in the role of interpreter.

---

Context

What this skill handles: Interview transcripts, survey open-text responses, support tickets, NPS/CSAT comments, session recording notes, usability test observations, mixed sources. The core rule: AI clusters and retrieves. You interpret. Never ask AI to tell you what findings mean — ask it to show you the evidence and you decide.

---

Step 1 — Receive and inventory the research

Gather: research objective, sources included, time period, and known context.

Step 2 — Prepare the data

For transcripts: extract participant speech only, label with codes (P1, P2). For surveys: treat each response as a unit. For tickets: extract core need, strip identifying info. Mixed: label every piece with source type.

Step 3 — First pass: tag without clustering

Tag every meaningful unit with descriptive, specific, data-derived topic tags. No themes yet — only tags.

Step 4 — Second pass: cluster into themes

Group by tag co-occurrence. For each theme: label (descriptive, not a conclusion), tag cluster, evidence count, sources, 3–5 supporting quotes (exact), and contradicting evidence.

Theme naming rules: observations not conclusions, never name as a solution.

Step 5 — Third pass: weight and rank

Assess each theme on Frequency (High >50% / Medium 25-50% / Low <25%) and Intensity (emotional language and emphasis level). Produce ranked priority table.

Step 6 — Fourth pass: identify contradictions and gaps

Document contradictions with opposing quotes, possible explanations, and implications. Document gaps with what's unknown, why it matters, and how to fill it.

Step 7 — Produce the synthesis report

Full report with: priority themes (evidence + interpretation clearly separated), weak signals, contradictions, research gaps, recommended next steps, and what the research does NOT tell us.

Quality check before delivering

Every theme has at least 3 supporting quotes — exact, not paraphrased
Contradicting evidence is included where it exists
Themes are named as observations, not conclusions or solutions
Interpretation is clearly separated from observation
Frequency and intensity are assessed separately
Contradictions and gaps sections are present
"What this does NOT tell us" is honest and specific
Suggested next step: Share the "Product implication" lines (not the full report) with your engineering team in sprint planning. Keep the full report as the evidence base.