Not every response is equally valuable. ChatWisp's Trust Engine automatically scores the quality of each completed session — reliably and with a clear explanation — so junk does not quietly distort your results.
What the Trust Engine watches for
- Speeding — answers arriving faster than the question could possibly be read.
- Straight-lining — giving the same rating to every question.
- Gibberish or low-effort text — keyboard mashing, "asdf," "idk," or one-word answers to open questions.
- Duplicate answers — verbatim responses copied across multiple sessions.
- Repair thrash — the interviewer having to re-ask many times because answers were hard to parse.
- Contradictions — a high recommendation score alongside clearly negative comments.
Quality levels and your control
Each session gets a label (high / medium / low / flagged) with a plain-English explanation. Flagged responses are set aside from your charts by default — so they do not skew your averages — but your raw export always stays complete.
You are fully in control: open any response and Accept it (to count a flagged one) or Reject it (to exclude one the engine missed). Charts recalculate accordingly, and you always see what changed.
Trust your numbers
The Trust Engine does not delete or hide data — it gives you transparency and control. Low-effort and bot-like responses do not silently distort your KPIs, and you have the final say on every judgment call without ever losing the underlying data.
Common questions
What counts as a low-quality response?
Speeding through questions, straight-lining the same answer down a scale, gibberish text, and verbatims duplicated across sessions. Each is scored separately, so one odd signal does not condemn an otherwise good response.
Does ChatWisp delete bad responses automatically?
Never. Flagged responses go to a review queue for you to accept or reject, and your decision is recorded. Nothing is removed on your behalf.
Will rejecting a response remove it from my export?
No. The raw export always stays complete so your data is never silently trimmed. Rejection affects the quality-weighted charts, not the underlying record.
Do AI interviews get less junk than static surveys?
They tend to, because a follow-up question is harder to satisfice through than a grid of radio buttons. The Trust Engine still runs either way.