Methodology
What the interviewer may do, what it may not, and how to check.
Adaptive follow-ups are useful only if you can trust and audit them. This page documents the controls in Candor, the risks that remain, and the parts of your data that stay comparable across respondents.
Last reviewed October 12, 2026
01 · Controls
Bounded by design
- 01
You write the anchors
Your authored questions set the order, with translation when another respondent language is selected. Follow-ups are inserted between them, never in place of them.
- 02
Depth has a ceiling
You choose the follow-up depth. The conversation budget includes any bonus follow-ups you enable; follow-ups are not separately metered or billed.
- 03
It stops when the answer is enough
If an answer is already specific, the interviewer moves on instead of filling a quota of questions.
- 04
You can restrict topics
You can name subjects the interviewer should not raise. Automated checks reduce risk, but they do not guarantee perfect compliance.
- 05
Respondents can skip
A short or declined answer is accepted. No one is cut off mid-survey, and partial answers are saved.
- 06
Every word is kept
Themes and summaries link back to quotes and full transcripts so a reader can check them.
Detailed behavior: How adaptive follow-ups work
02 · Comparability
Scores stay fixed. Follow-ups are qualitative.
An NPS question asked with the same wording and scale is as comparable as it was before you added follow-ups, provided it still comes first. The follow-ups that come after it are different for each person by design.
That means you should not count follow-up answers as if they were responses to one standard question, or compare them directly with an older open-text field. Use them the way you would use interview notes: to explain the number, not to replace it.
Comparable across respondents
- Score and rating questions
- Closed choice questions
- Anchor questions asked as written
- Trend lines over time, if wording is unchanged
Qualitative evidence only
- Follow-up wording and answers
- Number of follow-ups per person
- Theme counts, as an approximate guide
- Comparisons with a legacy comment box
03 · Risks
What can still go wrong
- Leading questions
- A follow-up can suggest an answer ("was it the price?"). Style guardrails check wording before it is shown, and you should still read a sample of transcripts from every study.
- Uneven depth
- Talkative respondents may receive fewer follow-ups because their first answer was already specific. Do not read follow-up count as a quality signal per person.
- Summary drift
- A theme label can overstate what people said. Treat themes as an index into quotes, and open the transcript before acting on a theme.
- Who responds
- Adaptive follow-ups do nothing for non-response bias. The people who answer a churn survey are not all the people who churned.
- AnswerI stopped using it after a few weeks.
- AskedLeading: Was that because it was too complicated?
- AskedNeutral: What changed around that time?
Written for this page. The first suggests a cause; the second lets the respondent supply one. Guardrails aim for the second, and transcript review is how you confirm it.
04 · Fit
When a human interviewer is the right call
Short adaptive interviews fit
- Recurring NPS, churn, and win/loss programs
- Many respondents, a few focused questions
- Teams without time to schedule calls
- Respondents in several languages
Book a human researcher for
- Sensitive, distressing, or high-stakes topics
- Early discovery where you do not yet know the questions
- Observing behavior, screens, or environments
- Complex buying committees and strategic accounts
05 · Evidence
What we have not measured yet
We have not published results showing that adaptive follow-ups produce more actionable answers than a static comment box. Any claim like that needs a controlled study with real participants.
The evaluation we intend to run, with consenting design partners:
- 01Same core question to comparable groups: static open text, scripted follow-ups with a matched question budget, and bounded adaptive follow-ups.
- 02A rubric for an actionable reason, defined before data collection, applied by blinded human coders.
- 03Measures: share of responses with a concrete cause, completion rate, completion time, leading-question incidents, and whether themes are supported by the original answers.
- 04Longer answers alone do not count as better. Results published with sample size, uncertainty, and limits, using anonymized aggregates only.
Status: proposed. No participants enrolled and no results as of the review date.
Download the proposed pilot protocol06 · Data
Your respondents' answers
Transcripts and reports stay inside your workspace. We do not publish customer responses as marketing material. For what we collect and how it is handled, read the privacy policy. This page makes no certification or compliance claims.
Read your own transcripts before you trust anyone's
Run a small study on the Free plan and check every follow-up yourself.