≈598 tokens on demand. Your agent reads this file only when SKILL.md points to it.
AI-Assisted Research & Synthetic Users
Purpose: Define where AI helps research, where it does not, and how to use synthetic users safely.
Contents: current adoption signals, AI role split, synthetic-user guardrails, decision flow, risk checklist.
Current Signals
| Signal |
Value |
| Researchers citing AI analysis/synthesis as a top trend |
88% |
| Researchers seeing synthetic users as impactful |
48% |
| Time reduction from AI-assisted qualitative analysis |
up to 80% |
| Synthetic agents mimicking survey answers in some studies |
up to 85% |
Role Split
- AI handles
what: transcription, clustering, coding suggestions, draft structure.
- Humans handle
why and what to do: empathy, judgment, ethics, prioritization, strategic interpretation.
Synthetic Users: Use And Non-Use
| Do |
Don't |
| Use for preparation or hypothesis generation |
Use as a replacement for real-user research |
| Treat outputs as hypotheses to validate |
Present synthetic output as user evidence |
| Use with widely documented user groups |
Use with niche or specialized populations |
| Pilot interview guides or proto-journeys |
Use for final concept validation |
Decision Flow
Use synthetic users only when all are true:
- the target group is broadly documented
- the task is preparation, desk research, or early hypothesis generation
- stakeholders are unlikely to mistake it for real research
- a real-user validation step is planned
Otherwise: use real-user research.
Risk Checklist
| Risk |
Mitigation |
| Hallucinated patterns |
Verify against source data |
| Bias amplification |
Use diverse source material and bias checks |
| Privacy leakage |
Anonymize data before AI processing |
| Skill atrophy |
Keep periodic manual analysis in the workflow |
| Loss of human nuance |
Keep interviews and final interpretation human-led |
Field Usage By Phase
| Phase |
Acceptable AI use |
Hard boundary |
| DEFINE |
Summarize prior research |
Verify summary accuracy |
| DESIGN |
Pilot guides or synthesize assumptions |
Do not skip human study design review |
| ANALYZE |
Assist coding and pattern detection |
Validate against source data |
| SYNTHESIZE |
Draft structure |
Human owns final narrative |
| DISTILL |
Trend summarization |
Human owns calibration |
