Open-Ended Survey Questions: Examples for Every Research Goal

Most insights teams run surveys for breadth and interviews for depth. The open-ended question is the hinge between them: it is where a survey tries to get a why, and where it usually collects “fine,” “n/a” and “nothing.” This library of copy-ready open-ended survey questions, grouped by research goal, makes those questions work harder, and marks the point where a question belongs in an AI-moderated interview instead.

Open-ended survey question examples by goal

Each set is written exactly as it would appear in a survey. Teams can paste a set into a questionnaire and swap in the product, brand or category name.

Customer satisfaction (CSAT) follow-ups

Use these directly after a satisfaction rating, and route by score where the survey allows it so unhappy customers see the “fell short” questions. When the top “fell short” answer is one vague word, that is the question to ask in an interview.

• What is the main reason for the rating you just gave?
• What was the best part of your experience today?
• What, if anything, fell short of what you expected?
• What is one thing we could have done to make this visit better?
• Which part of the process took longer than you expected?
• If you contacted our support team, what happened when you did?
• What would you tell a friend about booking with us?

NPS follow-ups

Pair these with the 0 to 10 likelihood-to-recommend question. The first question is the standard follow-up Bain & Company recommends. The rest work best routed to promoters (9–10), passives (7–8) or detractors (0–6). A detractor who writes “support” has a story a survey cannot ask for.

• What is the primary reason for your score?
• What would need to change for you to rate us a 10?
• (Promoters) If you were recommending us to a colleague, what would you say?
• (Passives) What is holding you back from recommending us more strongly?
• (Detractors) What is the biggest problem we need to fix?
• What do you compare us to when you think about recommending us?
• Who would you not recommend us to?

Product feedback and discovery

Use these in post-launch pulses and discovery surveys, when the team needs to learn how a product fits into someone’s week. Workarounds are easier to show than to type, and an interview lets participants walk through them.

• What do you use [product] for most often?
• Think about the last time you used [feature]. What were you trying to get done?
• What do you still have to do outside [product] to finish that job?
• Which feature would you miss most if it disappeared tomorrow?
• What did you expect [product] to do that it doesn’t?
• What almost stopped you from using [product] in your first week?
• Describe a recent moment when [product] saved you time.

Purchase and switching decisions

Send these after a first purchase, upgrade or switch, while the decision is still fresh in the respondent’s memory. The options considered and why they lost are a conversation, not a text box.

• What was happening that made you start looking for a [category]?
• What other options did you consider before choosing us?
• What made you choose us over those options?
• What nearly made you choose something else?
• Where did you first hear about us?
• What made you stop using the product you had before?
• What did others in your household or team need to hear before you signed up?

Brand perception and messaging

Use these in brand tracking and message tests. Show the message or ad on screen before asking. Three-word associations tell the team what; an AI-moderated interview asks what each word means.

• What three words come to mind when you think of [brand]?
• In your own words, what is this message saying?
• What in this message, if anything, is hard to believe?
• Who do you think this message is meant for?
• What would you expect from a company that says this?
• How is [brand] different from other [category] options you know?
• What would you do next after seeing this ad?

Concept and idea testing

Ask these after showing a concept description, image or prototype, and before any rating scale that might anchor the reaction. “Confusing” needs a follow-up: confusing where, and compared to what?

• What is your first reaction to this idea?
• In your own words, what does this product do?
• What, if anything, is unclear or confusing?
• Who do you think this is for?
• What would you use this instead of?
• What would make you hesitate to try it?
• What would you want to know before buying it?

Pricing and value

Use these alongside price-perception scales. They explain the judgment behind a number without putting a figure in front of the respondent. Value judgments unfold in steps that a single answer box flattens, and an interview can hear each step.

• How do you decide whether [product] is worth what you pay for it?
• What do you compare our price to?
• What would make the price feel fair to you?
• Which part of [product] feels most worth paying for?
• What, if anything, feels like it costs more than it should?
• What would you expect to get if you paid more?
• When did [product] last feel like good value to you?

Churn and cancellation

Use these in exit surveys at the moment of cancellation, and in win-back surveys sent within a few weeks. If the top answer is “too expensive,” move that question into an AI-moderated interview.

• What is the main reason you cancelled?
• When did you first start thinking about cancelling?
• What, if anything, could we have done to keep you?
• What are you using instead, if anything?
• What will you miss about [product]?
• What would need to change for you to come back?

Website and app experience

Run these as an intercept after a session or task, while the visit is still fresh. A blocked task is best narrated while the participant retraces it, which a conversation allows and a text box does not.

• What did you come to do today?
• What, if anything, got in the way of doing it?
• What was confusing or hard to find?
• Where did you expect to find [item]?
• What almost made you leave the page?
• What is one thing you would change about this page?

  The same question, both formats    

An illustrative exchange, not drawn from a real study. The library question asks what made a participant stop using their previous product. The survey records one answer. The interview can follow up on it.

Survey answer: Price.

AI-moderated interview:

Moderator: What made you stop using the product you had before?

Participant: Price.

Moderator: What changed about the price?

Participant: The plan went up the same month the TV app started buffering every night.

Moderator: Which of those would have been enough on its own?

Participant: The buffering.

The survey recorded a pricing problem. The interview found a product problem.

What makes a question open-ended

Researchers use open-ended questions to collect respondents’ own words in an empty answer box with no options. With closed-ended questions, respondents select fixed choices or a scale rating, and researchers get a number the team can track over time. The two formats differ on each of these points:

When to use an open-ended question instead of a scale or multiple choice

Open-ended questions cost respondents more effort than any other question type, so each one should cover ground a closed question cannot. Four situations justify the cost:

• The answer options are unknown: A fintech team asking why customers opened a second checking account elsewhere has no reliable list of reasons yet. An open question collects them, and next quarter’s survey can turn the top five into multiple choice.
• A score needs a reason: A travel brand that sees CSAT drop on one route needs to know what changed. The rating locates the problem, and the open-text answer names it.
• The research is early: In discovery work, the team is still learning the words customers use. Verbatim answers supply that language for later closed questions and for messaging.
• A rating or NPS question has been answered: The respondent has already formed a judgment, so a follow-up asking for the reason gets a sharper answer than a general comment box at the end.

If the team can write the answer options from memory, use multiple choice. If it can’t, or if a number needs a reason, ask an open question.

How to word open-ended questions: weak vs strong rewrites

Most one-word answers trace back to the question. Vague prompts get vague replies, and leading or double-barreled wording distorts what comes back. Hypotheticals ask people to predict their own behavior, which they do poorly. Questions about a recent, specific experience work better because a shorter recall window is easier to remember accurately. The rewrites below cover each failure:

The same fixes apply to discussion guides. The article on writing unbiased interview guides covers leading, loaded and double-barreled questions with more before-and-after examples.

How many to include and where to place them

Most consumer surveys need two or three open-ended questions, each tied to a decision the team has to make. A post-purchase survey, for example, might ask one CSAT follow-up and one question about what nearly stopped the purchase, then stop. Each extra text box costs effort, and respondents pay for it with shorter answers on every question that follows.

The cost grows with survey length. In a 2009 Public Opinion Quarterly study, Galesic and Bosnjak found that the longer a web survey’s stated length, the fewer people started and finished it. Answers to questions placed later in the questionnaire were also faster, shorter and more uniform.

Placement and requirement settings follow from that:

• Put each open question right after its rating: The respondent’s reasoning is still in mind, and the answer stays on topic.
• Keep earlier closed questions in mind: Related closed items asked first tend to shape which reasons people mention, so a pure discovery question belongs near the start.
• Make most open questions optional: Forcing an answer raises dropout and fills the dataset with “n/a” and keyboard mashing. Requiring one question that drives a decision, such as the cancellation reason, can be worth the trade.

If a question will not change a decision, cut it. If it will, place it right after the rating it explains.

Designing the answer box

Respondents read the size of the answer box as an instruction. A one-line field tells them a few words will do; a box with room for several lines tells them the team wants an explanation. Larger boxes draw longer answers, and the effect is strongest among the least motivated respondents, who are the ones most likely to type a single word.

Tight character limits cut answers off at the point where they get useful. In Strella data from 31 research participants who had recently taken part in both AI-moderated interviews and open-ended surveys, one participant described hitting a limit: “I was actually getting close to the character count limit so I just it all out and gave a shorter answer to stop wasting my time.” That answer records how much the participant was willing to type, which is less than what they thought.

Mobile needs its own check. Answers typed on phones run shorter than answers typed on computers, and a box that looks generous on a desktop can shrink to two lines on a small screen. Teams can use a full-width box that grows as the respondent types, preview every open question on a phone before launch, and put “(optional)” in the question label so respondents know they can skip without hunting for a button.

How to analyze open-ended responses

Open-text answers become decision-ready once a team builds a coding frame, tags every response against it and counts the results. The process runs in five steps:

1. Read a sample of answers and draft codes from what respondents wrote, adding codes the research question requires.
2. Write a codebook with a label, a one-line definition and an example quote for each code.
3. Tag every response, allowing more than one code per answer.
4. Count codes and state the denominator, because multi-coded answers push totals above 100%.
5. Pull two or three representative quotes for each major code.

On larger studies, two coders tag the same sample and compare results before the codebook is final. Disagreements usually mean a definition is too loose.

A streaming service’s exit survey shows how this works at small scale. Twenty cancelled subscribers answered “What is the main reason you cancelled?” A few of the answers:

• “Price went up again and I’d finished the show I signed up for.”
• “Buffering every night on the TV app.”
• “Switched to the one my roommate pays for.”
• “Nothing new I wanted to watch since spring.”

Five of the 20 answers carried two codes, giving 25 codes in total:

Price leads, but reading the seven price quotes splits it in two: some respondents objected to the increase itself, while others said the service was “not worth it for how little I watch.” The first group points to a pricing decision and the second to an engagement problem, and a count alone would have merged them.

A word cloud of the same 20 answers would have shown “price” and “show” in large type and stopped there. Word clouds count words instead of reasons, and most tools strip “not” as a stopword, which turns “not worth it” into “worth it.” They rarely support a decision.

The same coding-frame logic carries over when open-ended answers come from interviews instead of a survey. Strella, an AI-moderated interview platform, auto-codes open-ended answers from interviews into themes by question and objective and charts them, for the questions every participant was asked. Follow-up probes that only some participants received stay qualitative, and every theme links to the participant’s own words and video. Those counts describe the people in the study. A theme six participants raised is a lead for the team to check, and it does not translate into a market share.

Where surveys stop and interviews start

Surveys are the right call for quick pulses, broad reach, trend metrics and statistical analysis. A monthly CSAT tracker across a large member base needs a survey, and no interview program replaces that. The limit is the follow-up. When a respondent types “too expensive,” a survey cannot ask “compared to what?” and it moves on to the next question.

People also type less than they say. The Strella participant data above put the difference in participants’ own words; one said, “You could say a lot more in an AI interview than you can type, without a doubt.” Participants in the same study ranked the formats they prefer:

• Only 5 of 31 picked surveys as their favorite.
• AI-moderated interviews tied with human-moderated interviews for the top spot.
• Participants also credited surveys with strengths: they are faster, self-paced and reach more people.

AI-moderated interviews are their own methodology, sitting between surveys and in-depth interviews. An AI moderator holds a voice-to-voice conversation with real people, follows the researcher’s discussion guide and generates each follow-up from what the participant just said. On Strella, teams can run the same questions from their survey as voice-to-voice conversations.

Human-moderated interviews are live on the same platform, and teams choose AI or human moderation per study; human moderation is the right choice where cultural nuance or emotional weight is high. To hear how the moderator follows up, teams can try an interview.

Ritual used this approach on pricing. As participants moved through checkout, Strella asked questions and dynamic follow-ups on perceived benefits, value and pricing comprehension, and the resulting pricing work drove a 24% increase in net revenue per user. The comparison of AI-moderated interviews vs surveys covers what each format tells an insights team.

The two formats work best as a pair, with the survey for breadth and Strella for depth in qualitative research. If the team needs to know how many, keep surveying. If the next question is why, move it into an interview.

Have a survey result that needs a why? Bring your research question to us and we will talk it through. See Strella in action

FAQ

What are good open-ended survey questions?

Good open-ended survey questions ask about one topic, refer to a recent and specific experience, and avoid suggesting an answer. “What is the main reason you cancelled?” and “What nearly made you choose something else?” work because a respondent cannot answer them with yes or no.

What is the difference between open-ended and closed-ended questions?

Use closed-ended questions to track a number over time and open-ended questions to learn the reasons behind it. A survey often pairs them, with the open question placed right after the rating it explains.

How many open-ended questions should a survey have?

Most consumer surveys need two or three open-ended questions. Each one should map to a decision the team needs to make. Each extra text box costs respondents effort and shortens their answers on every question that follows.

How do you analyze open-ended survey responses?

Analyze them by building a codebook, tagging each response with one or more codes, counting the codes and pulling representative quotes. State whether percentages are based on respondents or on codes, and read the quotes behind the largest code before reporting it, since one label can hide two different problems.

What should follow an NPS question?

An open-ended “What is the primary reason for your score?” should follow the NPS question, placed directly after the rating. Teams that can route by score add a second question tailored to promoters, passives or detractors, such as “What is the biggest problem we need to fix?” for detractors.

How do you get longer answers to open-ended questions?

Teams can encourage longer answers by asking specific questions about recent experiences, providing larger answer boxes and avoiding tight character limits.  When teams need a follow-up that a survey cannot provide, an AI moderator can ask it in conversation.