By Vellora Labs · Reviewed October 10, 2026
Use this guide to explore how someone understands a concept and whether it connects to a recent task. It is designed for a human-moderated pilot. The suggested 25-minute structure is a planning aid, not a validated standard or a sample-size recommendation.
Before the first session
Write the product decision, audience, stimulus version, intended meaning, and factual limits. Confirm consent, recording permission, data handling, compensation, and accessibility arrangements. Replace the bracketed instructions in the download. Do not promise conditions that are not part of your actual study.
| Time | Purpose | Prompt or action |
|---|---|---|
| 0–3 min | Introduce the study | Explain purpose and consent; make it clear that disagreement is useful |
| 3–7 min | Ground the context | Ask for the most recent relevant task and what happened |
| 7–9 min | Present the stimulus | Show the material without pitching its benefits |
| 9–14 min | Capture unaided understanding | Ask for an explanation in the participant’s own words |
| 14–20 min | Investigate fit | Return to the real task and explore what would help or prevent use |
| 20–25 min | Locate confusion and close | Replay after the first response; clarify, then explain next steps |
The core questions
- How would you describe this idea to someone who has not seen it?
- What would you expect to provide or do first?
- What do you believe the product does, and what would remain your responsibility?
- Thinking of your recent task, where, if anywhere, might this fit?
- What would you continue doing the same way?
- What evidence would you need before trying it?
Follow the answer with “What led you to that?” or a request for a specific example. Avoid naming the benefit you hope they noticed. Give participants time to answer without filling the silence with an explanation.
What the download records
The observation log keeps stimulus version, timestamp, question, exact words or actions, interpretation, and alternative explanation together. It distinguishes an unaided response from an answer after replay or moderator explanation. Use participant codes and keep identifying details out unless necessary for the study.
Handling a difficult moment
The participant asks what the product really does.
Record the question and what remains unclear. Invite their current interpretation before explaining. If you do explain, mark later answers as assisted; do not overwrite the initial response.
The prototype or video fails to load.
Record a delivery problem. Repair access or reschedule. A failed stimulus is not evidence that the participant misunderstood the concept.
The participant says they love it.
Ask what part of a recent task it would change and what they would need before trying it. Keep liking, comprehension, and observed behavior as separate evidence.
Finish with a decision, including its limits.
Review the observations and contradictions together. Record the proposed revision, who will own it, and what you still do not know. Do not convert a handful of sessions into a population success rate or a claim of product-market fit.
Prepare recruitment with the screener builder. For other goals, use the question builder. For test delivery, explore the concept-testing tools collection.
This is an original Vellora template, not participant evidence. It can be adapted for AI moderation only after a pilot confirms that the actual system respects consent, probing boundaries, and the distinction between unassisted and assisted answers.