Writing Open-Ended Questions That Produce Classifiable Answers¶

The quality of a Text Analytics category set depends on the quality of the responses feeding it. A vague question tends to produce vague, one-word answers - which are exactly the hardest kind to classify meaningfully (see Qualitative Data Quality: What Makes a Response Hard to Classify). A few wording habits reliably produce richer, more classifiable answers.

Ask About Something Specific¶

"What did you think of the product?" invites a one-line reaction. "What specifically would you change about the onboarding process?" invites a substantive answer with real content to classify. The more specific the prompt, the more likely respondents are to give an answer with actual themes in it, rather than a generic reaction.

Explain Why You're Asking¶

A brief note on what you'll do with the answer ("so we can prioritize what to fix first") tends to produce more thoughtful responses than a bare question with no context. Respondents who understand the question matters are more likely to put real effort into answering it.

Avoid Leading or Double-Barreled Questions¶

A question that bundles two things at once ("What did you think of the pricing and the onboarding?") produces answers that are hard to classify cleanly, since a respondent may only address one half. If you need both, ask them as separate questions.

Give Enough Room to Answer¶

A text field that feels too small can unconsciously signal to a respondent that a short answer is expected. Make sure your open-ended fields don't visually suggest brevity if you're hoping for detail.

Consider AI Follow-up for Thin Answers¶

If a respondent gives a short or vague answer, AI Follow-up Questions can probe for more detail in the moment rather than leaving you with an unclassifiable one-liner after the fact - see AI Follow-up Questions: What They Are and Where They Work and AI Follow-up Questions: What Triggers One and What Doesn't for how to set a trigger condition aimed at exactly this case.

Placement in the Survey Matters¶

An open-ended question asked early, before respondent fatigue sets in, tends to get more thoughtful answers than the same question asked as the last item in a long survey. If a specific open-ended question is important to your analysis, consider where it sits in your survey's flow.

Test the Question Yourself¶

Before launching, answer your own open-ended question honestly as if you were a respondent. If your own answer comes out short or vague, respondents are likely to respond the same way - that's a sign to rework the wording before you collect real data against it.

FAQ¶

Does question wording affect AI category generation, or just response quality?
It affects response quality directly, which in turn affects what category generation has to work with - richer responses give the generation step more substance to draw real themes from.

Is there an ideal character limit for open-ended answers?
Not a fixed one - the goal is enough room that a thoughtful respondent doesn't feel cut off, without demanding more length than the question actually warrants.

Should I always pair an open-ended question with a structured one on the same topic?
It's a common and useful pattern - a structured question gives you a clean measure, and the paired open-ended question gives you the "why" behind it, which Text Analytics can then help make sense of at scale.

Can I fix a poorly worded question after I've already collected some responses?
Editing the question going forward is possible, but responses already collected under the old wording won't retroactively improve - if response quality has been consistently poor, consider whether a mid-survey wording change is worth the inconsistency it introduces.

For what to do once you've collected responses, even imperfect ones, see How Text Analytics Works: From Open-Ends to Categories.