The AI interviewer
Quality guardrails
Every AI question passes a four-rule check before a respondent sees it. You can monitor it all.
Before any follow-up reaches a respondent, a separate validation model checks it against four rules:
- Neutral: no leading or loaded phrasing.
- Single: exactly one question is being asked.
- Grounded: nothing invented beyond the respondent's words and your context facts.
- Boundary: nothing that touches your restricted topics.
A failed question is regenerated once, with the validator's feedback; if it fails again or takes too long, the conversation advances to your next anchor instead. The respondent never hangs and never sees a rejected question.