How adaptive follow-ups work
After each answer the AI decides to probe, advance, or end. Your study's follow-up depth sets the budget per conversation, with a small bonus reserve for high-value moments.
After every answer, the interviewer makes one decision: probe (ask a follow-up), advance (move to the next anchor), or end the conversation. It probes only when the answer contains something your objective cares about, within your study's follow-up depth budget. Short answers get special attention: when someone names a problem or idea in a one-liner (15 words or fewer), the interviewer is instructed to ask for a recent, concrete example rather than move on.
Choosing a follow-up depth
Set the depth per study in AI Configuration. The change applies to new conversations immediately; conversations already in progress keep the budget they started with. Respondents can also skip any question, and a skipped question never triggers a follow-up.
| Depth | Follow-up budget |
|---|---|
| None | 0 follow-ups, anchor questions only |
| Light | At most 1 follow-up per conversation |
| Standard (default) | At most 2 follow-ups per conversation |
| Deep | At most 4 follow-ups per conversation |
Changing the depth affects new conversations only. Interviews already in progress keep the budget they started with, so respondents never gain or lose follow-ups midstream.
Bonus follow-ups for high-value moments
When an answer carries an unusually strong signal, the interviewer can earn up to 2 bonus follow-ups beyond the depth budget for that conversation. Three signals qualify: a bold strategic idea worth treating as a mini-interview, a sensitive disclosure (for example burnout, equipment, or leadership), or a frustrated or flat-neutral mood whose problems have not been explored yet. Depth None is an explicit off switch: it never gets bonus follow-ups and never calls the AI.
What the AI is told to do
- Keep every question to 25 words or fewer, in a neutral tone.
- Ask exactly one thing at a time, no chained "and/or" questions.
- Stay grounded in the respondent's words and your context facts, never invent product details.
- Never ask about pricing you have not mentioned, or for personal information.
- Write in the language the respondent is taking the survey in.
Score-aware probing on NPS
| Score | Follow-up angle |
|---|---|
| 9 to 10 (promoter) | What specifically earned the score. |
| 7 to 8 (passive) | What is missing to make it a 9 or 10. |
| 0 to 6 (detractor) | What went wrong or fell short, asked plainly and without assumptions. |
Built for speed
While a respondent is choosing an NPS score, Candor quietly pre-generates a candidate follow-up for each of the three score buckets. When they answer, the matching question is usually ready instantly. Live generation gets a 3 second budget; if the model does not answer in time, errors, or its question fails validation twice, the conversation simply advances to your next anchor. Respondents never wait on a spinner or see an error from the AI.