Qualitative Survey Questions That Get Detail

A customer cancels a subscription, and the exit survey asks, “How can we improve?” The reply is “Price.” That answer may be sincere, but it gives the team little to work with. Was the subscription unaffordable from the start? Did its value become less clear after a particular change? Did the customer find another way to do the job? The survey collected a topic without collecting the experience behind it.

A better question would ask about the decision the customer just made: “What happened that led you to cancel today?” If the answer is still “Price,” a second question could ask what the customer expected to get for that price and what they actually used. The point of a qualitative survey question is to give people room to describe events, reasons, and language that fixed answer choices might miss. An open question allows a response in the person’s own words, although a written response can still be brief or leave out crucial context. Pew Research Center explains the distinction between open and closed questions.

The examples below are original, illustrative question wording, not tested survey items. Use them as starting points. The useful choice depends on whom you are surveying, what those people can reasonably remember, and what you will do with the answers.

Ask about an experience the respondent can actually recall#

“Tell us about your experience” sounds open, but it asks the respondent to choose a time period, an event, and a level of detail before answering. Some people will describe their most recent encounter; others will summarize a year. You cannot safely treat those answers as if they describe the same thing. Survey wording can shape answers, and clear, specific questions are easier to interpret. Pew Research Center discusses both wording effects and the need for specific questions.

Start by choosing an event that fits the decision you face. A restaurant deciding whether to change its online ordering flow needs accounts of recent orders, not general opinions about restaurants. An employer considering a new onboarding process needs accounts from people who joined during a defined period, not recollections of “starting work” across many years. A school trying to understand course registration needs to know which registration period the student is describing.

That decision produces a tighter prompt: “Think about the last time you placed an online order with us. What, if anything, took more effort than you expected?” The question identifies an event, leaves open the possibility that nothing went wrong, and asks for a concrete point of friction. If the survey follows immediately after an order, “the order you just placed” is clearer still. If the survey arrives months later, the respondent may not remember enough detail to answer confidently. A bounded period helps orient recall; it does not make memory perfect.

The word “qualitative” describes the kind of answer sought, not a magic opening phrase. “Why?” can be useful, but “What happened next?” or “What were you trying to do?” may bring the respondent closer to an observable experience. “Please explain” after a rating can also work when the rating has a clear subject. The strongest prompt names an object or moment and makes the desired description obvious.

Keep the research population in view. “What happened when you requested support in the past month?” assumes the respondent requested support. If that is not established, first ask whether they did, or phrase the question conditionally: “If you requested support in the past month, what happened?” The field period matters too. “This month” changes meaning if responses are collected over several weeks. Use a date range or an event named in the invitation when the timing needs to be consistent.

Customer and product questions that reveal the decision#

Customer surveys often ask people to rate satisfaction and then leave one empty box. That box is most useful when it asks about the decision the team can change. Suppose a subscription team wants to understand why new customers stop using a feature. “What do you think of the feature?” invites a verdict. “Think of the last time you needed to complete [task]. How did you do it, and where, if anywhere, did [feature] fit?” invites a description of the job, including the possibility that the feature was irrelevant.

For discovery, ask: “When you first looked for a way to [task], what were you trying to accomplish?” Follow with: “What other approaches did you consider before choosing this one?” Those questions separate the underlying need from the product name. A person might have compared a tool with a spreadsheet, a colleague’s help, or doing nothing. The alternatives in this sentence are illustrative possibilities, not claims about any particular market.

For purchase decisions, a useful pair is: “What information mattered most when you decided whether to buy?” and “What, if anything, was difficult to judge before you bought?” The first asks for criteria; the second asks about uncertainty. Avoid combining them into “What information mattered, and what was missing?” One answer may address only one half, leaving you unable to tell whether the other half was skipped or did not apply. Questions that combine two issues into one request create this interpretation problem; Pew Research Center calls them double-barreled questions.

For actual use, try: “Describe the last time you used [feature] to [task]. What did you do first, and what happened after that?” If the respondent may never have used the feature, screen for use before showing the prompt. For a self-administered survey, a separate text box asking “Where did you stop, if you stopped?” may be easier to answer than an invitation to reconstruct a long sequence in one box. The cost is more respondent effort, so keep only the steps that serve the decision.

For cancellation, ask: “What happened that made you decide to stop using [service]?” Then, if the survey allows a second open answer, ask: “What would have needed to be different for you to continue?” The second question is a counterfactual. Treat its answer as a stated possibility, not proof that a proposed change would have retained the customer. The first question remains the firmer anchor because it asks about the decision that occurred.

For a service recovery survey, “What happened during your most recent contact with our support team?” leaves room for both helpful and unhelpful experiences. “Which part of the response helped you move forward, if any?” can then identify a workable practice. “What should we change?” is often too broad on its own; a respondent can make a more useful suggestion after describing what happened.

These prompts serve different decisions. A team choosing product improvements needs descriptions of where a task broke down. A team deciding what to explain before purchase needs the uncertainty buyers faced. A team responding to cancellations needs the event and expectation that made leaving reasonable to the customer. Selecting a question by topic alone—“customer feedback”—hides those distinctions.

Employee, student, patient, and community examples need their own boundaries#

The same structure works outside a product survey, but the surrounding conditions change. An employee may worry that a detailed answer could identify them. A patient may describe private health information. A student may be reluctant to criticize a course while a grade is pending. Those possibilities affect whether the question should be asked, how much detail is appropriate, and what the invitation says about who will read the response.

For employee onboarding, an illustrative prompt is: “During your first two weeks in this role, what was one task you could not complete without help? What help did you need?” This asks for a defined period and an actual task. “How supported did you feel?” asks for an overall judgment; it might belong in the same survey as a rating, but it will not necessarily show which instruction, permission, or handoff needs attention. If you ask employees to name a colleague, make clear why that name is needed. A description of the process may be enough.

For a school’s registration process, ask: “Think about the course you most wanted to enroll in this term. What steps did you take, and where did you get stuck, if anywhere?” A companion question could be: “What information would have helped you choose among the available sections?” These prompts distinguish access problems from decision problems. A student who could not find a required course needs a different response from one who could not compare schedules, even if both answer a broad satisfaction question negatively.

For a healthcare service survey, focus on the service experience rather than inviting an assessment of medical care that the survey cannot interpret. An illustrative prompt is: “During your most recent appointment booking, what information did you need that was hard to find, if any?” If the research concerns explanations after a visit, ask: “Which part of the next-step instructions, if any, was unclear when you got home?” Give respondents a way to say the instructions were clear. If a service cannot protect or use detailed medical disclosures, do not ask a broad question that encourages people to enter them.

For a local consultation, ask: “When did you last use [named public service], and what were you trying to do?” Then: “What part of that visit or transaction should be easier, if any?” The service and period should be explicit. Otherwise, one respondent may discuss an online form while another describes an in-person encounter, and the team will have no clear basis for deciding which part of the service to change.

Sensitive identity questions require a separate design decision. An open box asking people to “explain your identity” may collect personal detail that the research does not need. If identity information is genuinely necessary, decide which concept the study needs, use wording and response choices suited to that purpose and population, and test whether respondents understand them. A UK review of sex and gender data collection illustrates how much the meaning and design of such categories matter. An experience question can usually stay focused on the experience without turning identity into an assumed cause.

Replace prompts that smuggle in an answer#

A poor qualitative question can be open in form while still steering the response. “How did our redesign make checkout easier?” presumes that checkout became easier. Someone who found it harder has to reject the premise before answering. Neutral wording gives that person a direct route: “What, if anything, changed about completing checkout after the redesign?” Neutral wording cannot repair a survey sent only to the happiest customers, but it removes one avoidable pressure within the question. Pew Research Center discusses how wording and survey design affect responses.

The rewrites below are illustrative. Each improved prompt still needs to be checked against the actual respondent group and the decision the survey is meant to inform.

The table is a set of design moves, not a promise that every right-hand question is universally best. For example, “hardest step” is a good prompt when the survey is intended to find friction. If the team needs a balanced account of what worked and what did not, first ask the respondent to describe the process and then ask which parts were easy or difficult. The research decision should control how narrow the question becomes.

Avoid jargon unless the respondents actually use it. A workplace question about “cross-functional enablement” may mean several things to different people. “When you needed information from another team last week, what did you need and how did you get it?” names an observable situation. Similarly, avoid asking people to explain a cause they cannot know. A customer can report that a page did not load for them; they usually cannot identify the technical cause from that experience. Ask for what they saw, what they attempted, and what happened next.

Build a short sequence, not a collection of attractive prompts#

A good standalone question can fail in a bad sequence. If a survey first lists possible complaints—cost, speed, staff behavior—and then asks “What mattered most?”, the list has already supplied ideas to the respondent. Pew Research Center notes that earlier questions can influence answers to later open questions. When the goal is to discover what comes to mind unprompted, place the broad open question before a list of proposed explanations.

An illustrative sequence for a service booking survey could begin with a simple filter: “Have you booked an appointment with us in the past three months?” People who say no should not receive questions about that booking. Those who say yes could then see: “Think about your most recent booking. What were you trying to arrange?” Next: “What happened from the time you started until you knew the appointment was booked?” Finally: “Which step, if any, would you change first?” This sequence moves from context to action to judgment. It asks for one booking rather than an average impression.

If the team also needs to know whether a person used the website, phone, or front desk, a closed question may be more efficient than another open box. Open questions let respondents introduce reasons and descriptions beyond fixed choices; closed questions can establish a known category quickly. The Open University’s discussion of open questions describes their value for drawing out experiences, while Pew Research Center explains the different answers that open and closed formats can produce. Use the format that fits each job instead of making every item open-ended.

Keep one clear request in each text box. “What were you trying to do, what went wrong, whom did you contact, and what would fix it?” looks efficient on the page, but a respondent may answer only the final part. A small number of focused prompts can produce a more interpretable record. The cost is space and effort: every additional box asks the respondent to recall, compose, and type. Choose the questions whose answers would change a decision, and remove those that merely sound interesting.

An optional “Anything else about this experience we should know?” can catch a detail the survey did not anticipate. It should come after the targeted questions and should not carry the main burden of the study. A generic final box cannot compensate for earlier prompts that never named the experience under study.

Test the words before the survey goes live#

The person writing a question knows what “support,” “access,” or “recently” is supposed to mean. The respondent does not share that private definition. A short pretest can reveal that one person reads “support” as technical help while another reads it as encouragement from a manager. Both answers may be intelligible in isolation and still be impossible to compare.

In a cognitive interview, ask a person from the intended population to answer the draft survey question, then ask what the key terms meant to them and what they recalled while answering. Pew Research Center describes using this approach to find unexpected interpretations and difficult questions. For the illustrative booking question, ask what counted as a “booking”: a completed appointment, an attempted booking, or a rescheduled visit. Revise the prompt or add a filter if those interpretations lead to different kinds of answers.

The pretest should also examine the response task. If people repeatedly ask whether they may discuss more than one visit, the question needs a clearer unit. If they can answer only after looking up a date or searching old messages, the recall period may be too long for the intended survey. If they answer with “fine” or “nothing,” ask whether the prompt is too broad, whether the subject is irrelevant, or whether the person genuinely had nothing to add. Do not treat every short answer as a writing failure.

A few interviews can show how people understood the wording; they do not show how common an interpretation is in the full population. Nor can pretesting remove every wording effect. Make the revision that resolves the observed confusion, then judge whether the changed question still asks about the same experience. For a survey repeated over time, even a small wording change may affect comparability. Pew Research Center discusses the importance of stable wording when tracking change.

Read the answers as accounts, then decide what can be counted#

Open-text answers do not arrive as clean categories. Imagine an illustrative set of responses to “What made booking difficult, if anything?” One person writes, “The button kept sending me back.” Another writes, “I could not tell which appointments were online.” A third writes, “No trouble booking; I just needed an evening slot.” Calling all three “website problems” would lose the distinction between a possible navigation failure, unclear information, and limited availability.

Before reporting themes, decide what each category means. A simple coding guide for that illustrative set might define “navigation” as trouble moving through the booking flow, “information” as missing or unclear details about an appointment, and “availability” as the absence of a suitable time. Let one response receive more than one code when it genuinely describes more than one issue. Preserve an “unclear” category for answers that do not justify a more specific label. Then reread borderline answers against the same definitions. This is a practical response to the interpretation work open questions create; The Open University describes how open questions invite varied experience-based answers.

Keep the respondent’s wording available alongside the codes. “The button kept sending me back” may point to a different next step from “I could not find the button,” even if both are initially grouped under navigation. A code is a way to organize what people wrote; it is not a substitute for reading the account or checking the product itself. Do not turn a plausible interpretation into a reported fact about the system without examining the relevant step.

Be exact about the denominator when describing a pattern. In an illustrative survey, suppose 200 people were invited, 80 submitted the survey, 50 wrote something in the booking text box, and 18 of those 50 mentioned appointment availability. “Eighteen of the 50 people who answered this text question mentioned availability” describes the observed text. “Thirty-six percent of customers could not find a slot” would overstate it: the 50 writers are only part of those invited, “mentioned availability” is not the same as “could not find a slot,” and the invitation itself may not represent all customers. The numbers here are simulated to show how the denominator changes the claim.

Silence is not a positive review. A respondent may skip a box because nothing happened, because they are tired, or because they do not want to write about it. A one-word response may be enough to establish a topic but not enough to identify a cause. If an answer is too thin for the decision at hand, a follow-up interview may help, provided the person has agreed to be contacted. A written survey is useful for collecting many accounts in a consistent format; a conversation allows a researcher to ask what a specific phrase meant.

When reporting what was learned, separate the answer from the inference. “Several respondents described uncertainty about which appointments were online” is a report about the text. “The booking page fails to label appointment types clearly” is a hypothesis to check against the page. “Adding a label will solve the problem” is a proposed intervention. The survey can point to the decision; it cannot, by itself, prove the intervention will work.

Choose the question you are prepared to use#

The best qualitative survey question is the one that gives a respondent a recognizable moment to describe and gives the reader of the answers a fair basis for action. For a purchase survey, that may be the information a buyer lacked. For a product survey, it may be the step where a task stopped. For an employee survey, it may be the handoff that delayed work. The words should allow a good experience, a bad experience, and an experience the designer did not anticipate.

Start with the decision. Name the relevant event and time frame. Ask one thing at a time in ordinary language. Place an unprompted question before any list of proposed explanations. Test the draft with people who resemble the intended respondents, then read the answers with clear interpretation rules. Each step costs time and some survey space, but it buys a more useful account than a generic invitation to “share your thoughts.”