Back
Open Ended Questions in Surveys: Getting Answers Worth Reading
Matthieu Saussaye
An open ended question asks respondents to answer in their own words instead of picking from a list. It is the only question type that can return something you did not anticipate. It is also the type most likely to come back as a single word, because typing costs effort and the respondent has no idea how much you want.
What determines whether open questions earn their place:
Use them to explain, not to measure. Closed questions tell you how many. Open questions tell you why, and only if they are written well.
Thin answers are a design failure, not a respondent failure. "Any other comments?" earns "no". A specific, time-anchored question earns a paragraph.
Keep the count low. A handful of open questions in a survey is workable. A wall of text boxes drives abandonment and one-word answers.
Place them after the closed question they explain. An open question in isolation gets a shrug. The same question right after a rating gets a reason.
Plan the analysis before you field. Open text becomes countable only after coding. If nobody is going to code it, do not ask it.
A follow-up probe usually beats an extra question. Asking "what do you mean by that?" recovers far more than adding a second text box further down.
Open ended versus closed ended, in practice
The two formats differ in what they cost you and what they can possibly return. This is the trade you are making every time you choose a text box over an option list.
Dimension | Open ended | Closed ended | What that means for you |
|---|---|---|---|
Who supplies the vocabulary | The respondent | You, in advance | Only open questions can surface something off your list |
Effort to answer | High: typing or speaking | Low: one tap | Open questions get skipped or answered in one word |
Effort to analyze | High: needs coding | None: already categorized | Analysis cost scales with sample size |
Comparability | Only after coding into themes | Immediate and exact | Percentages from open text depend on your coding frame |
Best used for | Reasons, wording, unknown unknowns | Sizing, tracking, segmentation | Use both; do not make one do the other's job |
The full breakdown of the fixed-answer side sits in the companion article on closed ended questions, and the wider framing is covered in qualitative versus quantitative research.
Why most open responses come back thin
Open ended answers of one or two words are the normal outcome of a badly framed question, not a sign that respondents have nothing to say. Four things drive it.
The respondent does not know what you want
A blank box gives no signal about length, topic or level of detail. Faced with ambiguity, people give the shortest answer that could plausibly count. "Good" is a rational response to "What did you think?"
Typing is expensive, especially on a phone
Most survey traffic arrives on mobile. A text box on a phone means a cramped keyboard, no visible context above it, and a real physical cost per word. The same person who would talk for two minutes will type six words.
The question asks for a summary, not a memory
"How would you describe your experience with our service?" asks the respondent to abstract across everything that ever happened. Abstraction produces adjectives. "Think about the last time you contacted support. What happened?" asks them to retrieve one event, and retrieval produces detail.
There is no consequence to answering briefly
In a conversation, a one-word answer gets a follow-up. In a survey form, it gets a Next button. Nothing in the interface communicates that a short answer was insufficient, so short answers persist.
How question wording changes answer depth
The difference between a useless open question and a productive one is usually a rewrite, not a new tool. Five moves do most of the work.
Anchor to a specific event. "Tell us about your experience" becomes "Think about the last order you placed. What, if anything, was frustrating about it?" One instance, one memory, real detail.
Ask what happened, not what they think. Behavior is easier to recall than opinion, and it is more useful. "What did you do when the payment failed?" outperforms "How did the payment failure make you feel?"
Presuppose that there is something to say. "What would you change first?" assumes there is a first thing. "Do you have any suggestions?" invites "no".
Narrow the scope. Broad questions get broad answers. "What could be better about the checkout?" beats "What could be better about the site?"
Avoid yes/no grammar. Any question starting with "do", "did", "is" or "would" can be answered in one syllable. Start with "what", "how" or "which part".
Two things that do not help as much as people expect: setting a minimum character count, which produces padding and abandonment rather than substance, and adding "please be as detailed as possible", which most respondents skim past.
How many open questions to include
There is no validated number, but the practitioner convention is consistent: few, and placed deliberately.
One or two in a short satisfaction survey. Usually a "why that score?" after the main rating, and one forward-looking question near the end.
Three to five in a longer study. Attached to the closed questions where the reason actually matters.
Never several in a row. Consecutive text boxes read as work. Quality drops sharply from the first to the second.
Place them after the closed question they relate to, not in a block at the end. The end of a survey is where attention is lowest, which is exactly why "Any other comments?" collects so little. If you keep only one open question, make it the one that follows your key metric, whether that is a Likert scale item or a single select question.
Making them optional is generally the right call. A mandatory text box converts people who had nothing to say into people who typed "n/a", and it pushes some of them out of the survey entirely, which feeds the wider problem of collapsing response rates.
Open ended survey template: 14 questions that get real answers
Use these as a questionnaire sample of open ended items. Each is written to trigger retrieval of a specific memory rather than a summary judgment.
After a rating or score
You gave a score of [X]. What is the main reason for that score?
What would have had to be different for you to give one point higher?
You said you were dissatisfied with support. What specifically went wrong?
Product and experience
Think about the last time you used the product. What were you trying to get done?
What is the one thing you would change first, if you could change only one?
Which part of the setup took longer than you expected, and what happened there?
What do you do today when the product cannot handle something you need?
Purchase and choice
What made you start looking for a solution in the first place?
What else did you seriously consider, and why did you not choose it?
What almost stopped you from buying?
Churn and non-use
What was happening around the time you decided to cancel?
You said you have not used it in the last month. What have you been using instead?
Language and positioning
How would you describe what we do to a colleague who has never heard of us?
When you read our homepage, what did you expect the product to do?
Question 13 is worth singling out. Asking people to explain your product in their own words is the cheapest positioning research available, and it consistently returns vocabulary that differs from your marketing copy.
The analysis problem, and how coding works
Open text is not data until it is categorized. Two thousand free-text answers are two thousand individual observations that cannot be counted, charted or cross-tabulated in that state. Coding is the process of turning them into something countable.
Build the coding frame from the data
Read a sample of responses first, typically a hundred or so, and write down the themes that actually appear. That list becomes your codebook. Building the frame from what you expected to hear defeats the purpose of asking an open question, because it re-imposes the closed question you were trying to escape.
Code every response against the frame
Each answer gets one or more codes. One answer can carry several: "it was expensive and the app kept crashing" is both a price code and a reliability code. Decide in advance whether you are coding multiple themes per response, because it changes how your percentages add up and what the base means.
Keep the frame stable, and keep the verbatims
If you are tracking over time, the codebook has to survive between waves or the trend is meaningless. Always keep the original text linked to the code, so any number in the report can be traced back to the sentences behind it. A theme you cannot illustrate with three real quotes is a theme you should re-check.
Report counts honestly
"22% mentioned price" means 22% mentioned it unprompted, not that 22% care about price. Unprompted mention is a weaker and different signal than a selected option, and it undercounts by design. Say which one you are reporting.
Automated coding
Language models can now do the first pass of theme extraction and assignment at a speed manual coding cannot match, which changes the economics of open questions substantially. The discipline stays the same: check the frame it produced against a sample you read yourself, and keep the verbatims attached. The failure mode is a tidy set of themes that nobody validated against the raw text.
When a follow-up probe beats another question
If a respondent answers "too expensive", adding a second open question later in the survey will not clarify it. They have moved on, and the second box gets a shorter answer than the first.
What recovers the meaning is a probe in the moment: "expensive compared to what?" asked immediately, while the thought is still active. This is the mechanism that makes interviews richer than surveys, and it is not about the interviewer being clever. It is about the follow-up arriving within seconds of the original answer.
A single well-probed open question generally beats three unprobed ones. It is also less work for the respondent, because they only have to think about one topic properly instead of three topics badly. The related read on scaling that up is getting more insight out of AI surveys, and the question-type map lives in the survey questions guide.
Ask the second question, not just the first
The reason open responses are thin is that a form cannot react. SmartInterview can. It runs surveys by voice or text with an AI that reads the answer, notices when it is vague, and asks the natural follow-up: what specifically, compared to what, what happened next.
Speaking also removes the mobile keyboard tax that shortens written answers, and every response is transcribed and coded into themes automatically, in whichever language the respondent used. You get counts you can put in a chart and the verbatims sitting behind each one.
Try it on your next open question: start free or book a walkthrough.
Frequently Asked Questions
What is an open ended questionnaire?
An open ended questionnaire is one built mainly from questions that respondents answer in their own words rather than by selecting from fixed options. It is used when the goal is to discover reasons, language and issues you have not anticipated. The trade-off is that the responses have to be read and coded into themes before they can be counted or compared.
What is an example of an open ended survey question?
"Think about the last time you contacted our support team. What happened?" is a strong example: it anchors to one specific event, asks for behavior rather than a summary judgment, and cannot be answered yes or no. A weak example is "Do you have any comments?", which invites a one-word answer and usually gets one.
How many open ended questions should a survey have?
Few, and placed after the closed questions they explain. One or two is typical for a short satisfaction survey; three to five for a longer study. Never place several in a row, because answer quality falls sharply from the first text box to the second. Make them optional so people with nothing to say do not type "n/a" or abandon.
Why do people give such short answers to open ended questions?
Because the question usually asks for a summary rather than a memory, gives no signal about how much detail is wanted, and costs real effort to answer on a phone keyboard. Nothing in a form reacts to a one-word answer, so there is no pressure to elaborate. Anchoring the question to a specific recent event fixes most of it.
How do you analyze open ended survey responses?
By coding them. Read a sample to build a codebook of the themes that actually appear, assign one or more codes to every response, then report the frequency of each theme. Keep the original text linked to each code so numbers can be traced back to real sentences, and keep the codebook stable if you are tracking across waves.
Are open ended questions qualitative or quantitative?
They are qualitative in what they collect and can become quantitative in how they are reported. The raw response is unstructured language. Once coded into a fixed set of themes, theme frequencies can be counted and cross-tabulated like any other variable, as long as you are clear that a mention count is not the same measurement as a selected option.
