AI Detectability

Two signals, checked together: a read on the transcript for the usual tells of a recited answer (missing filler words, a sudden shift to polished phrasing), and a check against local, reverse-engineered builds of the actual stealth copilots people use. Harder to fake than either alone. Scored per speaker, evidence included.

We run the actual copilots candidates use against real interview questions ourselves, in builds we've reverse-engineered — so it's a match against what one really produced, not a guess.

Genuinely zero friction — the bot joins your normal Zoom, Google Meet, or Teams link exactly as it already is. No secure meeting link to generate instead of your real one, no separate app, no camera, screen, or tab access requested from the candidate. As far as we've seen, Candit is the only AI Detectability option that works this way — any role, anytime, anywhere, in the meeting you were already going to have.

It's a signal to ask about, not a verdict. AI Assistance is flagged conservatively, and always has evidence attached to it.

Under ~40% is usually just good prep, not live reading. It's crossing 40% that's the real signal — closer to how a live copilot actually renders an answer, and worth an actual follow-up question.

Yes — the bot identifies itself by name when it joins, and every session is flagged as recorded. Nothing extra to disclose beyond that.

It's included on Plus, Pro, and Team — the one thing Free doesn't have.