Quality Controls
The review methods, AI moderation, and quality signals Terac gives you to keep low-effort and dishonest work out of your dataset.
Verification confirms who a participant is. Quality controls confirm that the work they submitted is genuine and meets your standards. Terac gives you several tools, from fully manual to AI-assisted, so you can match the level of scrutiny to the stakes of your study.
Choosing a Review Method
| Method | How it works | Best for |
|---|---|---|
| Manual review | You evaluate each submission yourself | Complex or nuanced work |
| Auto-judge | AI evaluates submissions against criteria you define | Standardized tasks with clear pass/fail criteria |
| Quality check | AI screens first, you review only the edge cases | High-volume studies that still need a human eye |
See Reviewing Work for the full workflow.
Decisions and Signals
When you review a submission, you have more than a binary choice. Each action feeds the signals that protect quality across the platform.
- Approve the work meets your standards and the participant is paid.
- Reject the work does not meet standards; you can include a reason.
- Strike flags serious problems such as repeated low effort or dishonesty.
- Participant ratings let you score quality after review. Ratings influence a participant's standing and whether they are recommended for your future opportunities.
- Kick out removes an unresponsive or spamming participant and frees their slot.
Because ratings and strikes carry forward, high-quality participants surface more often in your future studies and the panel improves over time. Reserve rejections and strikes for genuine quality problems.
AI Moderation and Verification
For studies that need deeper assurance than a form submission can provide, the AI moderator conducts voice interviews that screen, probe, and verify in real time. Beyond gathering qualitative depth, moderated interviews are well suited to verifying claims and surfacing inconsistencies on longer or higher-stakes tasks, where a static submission alone is easier to game.
You attach a research guide with a screening policy, and the moderator evaluates responses against it. See Research Guides and Interviews.
Attention and Comprehension Checks
Attention and comprehension checks are your first line of defense against low-effort responding inside a task. They work best when they are fair: they should test attention to instructions that genuinely matter, not trick attentive participants with ambiguous wording. Pilot your checks alongside the rest of your study, and treat a high failure rate as a sign that the check, not the participant, may be at fault. See Avoiding Bias.
A Layered Approach
No single check is sufficient on its own. The strongest setups combine:
- Required attestations so applicants are verified.
- Filters and screening so they fit your population.
- Fair in-task checks so low-effort work is visible.
- A review method matched to your stakes.
Quality assessment is about converging signals, not one number. Avoid rejecting on a single ambiguous indicator. Weigh completion behavior, check performance, and content together before acting.