Research Synthesizer

Research Synthesizer

Processes multiple transcripts or research notes and produces a structured thematic synthesis.

When to Use

• After cleaning transcripts (pairs well with the transcript-cleaner skill)
• When you have 5-15 transcripts or sets of notes to analyze
• For cross-study synthesis across related research efforts
• Before building stakeholder readouts or presentations

For sets larger than 15, run in batches of 10-12 and then synthesize the batch outputs together.

How It Works

Step 1: Ingest and Index

Read all provided transcripts/notes. For each document, extract:

• Participant identifier (use pseudonyms if real names aren't provided)
• Key statements, observations, and quotes
• Contextual metadata (role, company size, use case) if available

Step 2: Bottom-Up Theme Identification

Identify themes inductively from the data rather than imposing categories:

1. Extract atomic observations — individual claims, behaviors, or reactions from each participant
2. Cluster by similarity — group observations that address the same underlying topic or behavior
3. Name the clusters — give each theme a descriptive, specific name (not vague labels like "usability issues" — instead: "onboarding flow creates false confidence about feature completeness")
4. Assign prevalence — count how many unique participants contributed evidence to each theme

Step 3: Analyze Contradictions

This is where the most valuable insights often hide. For each theme, actively look for:

• Participants who expressed the opposite view
• Behaviors that contradict stated preferences
• Differences that correlate with participant attributes (role, experience level, company size)

Step 4: Identify Weak Signals

Flag observations from only 1-2 participants that seem significant despite low prevalence. These are often early indicators of emerging patterns that a larger study would confirm.

Output Format

Always structure the synthesis document as follows:

# Research Synthesis: [Topic/Study Name]

## Study Overview
- Number of participants: X
- Participant profiles: [brief summary of who was interviewed]
- Research questions: [what the study set out to learn]
- Date range: [when sessions occurred]

## Theme Summary
| Theme | Prevalence | Strength | Key Insight |
|-------|-----------|----------|-------------|
| [Name] | X/Y participants | Strong/Moderate/Emerging | [One sentence] |

## Detailed Themes

### Theme 1: [Specific, descriptive name]
**Prevalence:** X of Y participants
**Strength:** Strong / Moderate / Emerging

**Summary:** [2-3 sentences explaining the theme]

**Supporting Evidence:**
- P3: "[Direct quote]" — [context for the quote]
- P7: "[Direct quote]" — [context]
- P11: [Observed behavior or paraphrased insight]

**Nuances:** [Important variations within this theme]

[Repeat for each theme, ranked by prevalence]

## Contradictions
[For each contradiction, explain what's in tension, who said what, and possible explanations for the disagreement]

## Weak Signals
[Observations from 1-2 participants that seem worth investigating further, with reasoning for why they matter]

## Methodology Notes
[Any caveats about the analysis — sample size, participant skew, topics not covered]

Quality Guidelines

Do:

• Use direct quotes as evidence wherever possible
• Distinguish between what participants said vs. what they did (stated vs. revealed preferences)
• Note when a theme is driven primarily by one vocal participant vs. broadly supported
• Flag your confidence level for each theme

Don't:

• Editorialize or make product recommendations — that's the researcher's job
• Treat prevalence as the only measure of importance (a safety issue mentioned by 1 person can matter more than a convenience issue mentioned by 10)
• Merge genuinely different themes just because they're related
• Strip context from quotes in ways that change their meaning

Batch Processing for Large Studies

For 15+ transcripts:

1. Split into batches of 10-12 (group by participant similarity if possible)
2. Run synthesis on each batch
3. Create a meta-synthesis from the batch outputs, looking for:
   • Themes that appear across multiple batches (strong signal)
   • Themes unique to one batch (may correlate with participant attributes)
   • Contradictions between batches

Chaining with Other Skills

• Input from: transcript-cleaner (cleaned transcripts)
• Output to: stakeholder-readout-generator (synthesis becomes the basis for readouts), insight-tagger (themes inform taxonomy)