Last updated: 18 September 2026
Quick Answer: Sample open ended survey questions work when each one names a single moment and asks the respondent to describe it, not to rate it. Pew Research Center's audit of 92 open ends put average item nonresponse at 18 percent, against 1 to 2 percent on closed questions.
Two open ends on the same Pew questionnaire came back very differently: 3 percent skipped one on a preferred nominee, 52 percent skipped one on reducing inequality between Black and White people. The wording sets the bill.
What Open Ended Questions Cost in Completed Responses
Open ends come back thin because they ask the respondent to compose rather than choose. Pew Research Center's audit of item nonresponse put the average near 18 percent across 92 open ends, against 1 to 2 percent on closed questions, the only published audit of its kind. Two dimensions are yours to control:
- Requested answer length. Multi-sentence questions ran at 18 percent median nonresponse, against 13 percent for a phrase.
- Cognitive burden. Median nonresponse on medium-burden and high-burden questions ran two to three times that of low-burden ones.
The Survey Length Budget
Pew caps its online panel questionnaires at 15 minutes, pricing that cap at 85 points: a detailed open end costs 8, a straightforward one 5, a simple stand-alone closed question 1. Two or three open ended questions is the working limit: a fourth takes 8 points from closed questions the analysis plan may need. Place them before the battery that tires the respondent: skippers are disproportionately those for whom typing is expensive, so check who is left in the sample before quoting a verbatim as representative.
Question Order and Placement Effects
Where a question sits changes whether it gets answered, separately from how it is worded. A study of 27,212 respondents who had completed an alumni survey found 68 percent answered an open end near the beginning, 79 percent in the middle, and 24 percent near the end, though all had finished the entire survey. Fatigue outruns intent: an optional item near the end gets skipped anyway. Put the open end a study cannot afford to lose early or in the middle, never last.
What Makes an Open Ended Survey Question Work?
A working open ended survey question names one moment, asks for a description rather than a rating, and carries one pretested concept. Fail any of the three and you have bought nonresponse for data you could have tabulated.
- One moment, not a category. "What happened the last time you contacted us for help?" retrieves an episode. "What do you think of our support team?" asks for a summary nobody holds.
- Describe, do not rate. The open end explains the score beside it, so ask what happened, not how good it was. "How satisfied were you, and why?" is two questions wearing one number.
- One concept, pretested. The American Association for Public Opinion Research's best practices call for one concept per question, in words the audience uses, and the US Census Bureau's testing standard requires pretesting it, such as a cognitive interview where respondents describe their thoughts while answering a draft.
Open Ended Survey Question Examples by Research Goal
Thirty open ended questions for surveys, grouped by research goal. Square brackets are slots to fill.
General Feedback
- What were you trying to get done today?
- What nearly stopped you from finishing?
- What was the exact moment you knew this was, or was not, going to work?
- What did you expect to happen next, and what happened instead?
- What is the first thing you would fix if you were in charge for a day?
Customer Satisfaction
- What made you choose us over the alternative you were weighing?
- You rated us [score]. What put it there rather than one point higher?
- Describe the last interaction that nearly made you switch to someone else.
- What almost stopped you from buying the first time?
- What would have to happen for you to recommend us without being asked?
Product Feedback
- Describe the last time [product] did not do what you needed.
- What are you using instead of [feature] today?
- Walk me through the last time you almost gave up setting [product] up.
- What is the first thing you do in [product] every time, and why?
- What feature have you never opened, and what has stopped you from trying it?
Workplace and Employee
- What made your best day at work this month a good one?
- The last time you thought about leaving, what prompted it?
- Describe a moment this month you felt genuinely supported by your manager.
- What is one thing your team does that new hires are never told about?
- What would need to change for next year to be better than this one?
Education and Course
- What is one thing from this course you have already used?
- What did you expect this course to cover that it did not?
- Describe the moment a concept in this course finally made sense.
- What almost made you drop this course before the halfway point?
- What would you tell someone deciding whether to take this course?
Concept and Brand
- In your own words, what is this product for?
- What does [brand] stand for, in your own words?
- Who do you picture using this, and why them specifically?
- What is the first word that comes to mind when you see this, before anything is explained?
- What would make you trust this less?
Ask 26 and 27 before any rating scale: a check taken after a score shows what the respondent decided to defend. The same ordering governs an unaided brand awareness instrument.
Common Mistakes When Writing Open Ended Survey Questions
Four wording patterns account for most of the damage: asking two questions in one, burying a negative inside an agree-or-disagree frame, leading the respondent toward an authority's position, and offering an unbalanced set of options. Each is fixable once it is named.
- Double-barreled wording. AAPOR's guidance on question wording uses a voting question as its example: "Did you vote in the 2004 and 2006 elections?" cannot tell whether a "yes" means both years, 2004 only, or 2006 only. Split it into two questions instead.
- A double negative inside an agree-or-disagree frame. "On occasion, I am unable to express how interested in politics I am" leaves "agree" ambiguous between not interested and merely inarticulate. Rephrase in the positive: "I am usually interested in politics."
- Leading phrasing. A question that states an authority's position before asking, or supplies only one side of an issue, pulls the answer toward whatever it just supplied. State the issue plainly and offer both directions.
- Unbalanced response options. "All the time, most of the time, never" weights three of three options toward frequency. A balanced set holds equal room on both sides, midpoint stated or implied.
None of these are open-end-specific: they corrupt a closed question the same way, and a pretest catches them where a proofread does not.
How Do Open Ended and Closed Ended Questions Differ?
A closed question measures how many; an open end, or questionnaire open question, tells you why, in vocabulary you did not supply, and arrives as work rather than an answer.
Rows sorted alphabetically by the first cell.
| Instrument | What it measures well | What it costs | What it cannot do |
|---|---|---|---|
| Closed question with fixed options | Prevalence, change across waves, subgroup differences. | About 1 point in Pew's budget; 1 to 2 percent item nonresponse. | Surface a reason nobody put on the list. |
| Open ended question | Reasons, vocabulary, unprompted salience. | 5 to 8 points in the same budget; 18 percent average item nonresponse. | Produce a number without a coding step. |
What automated open-end coding does, and does not do. Software applies an existing code frame across verbatims, turning text into a distribution in minutes. Tested systems do not decide what the frame should contain, resolve verbatims between two codes, or report their agreement rate against human coders.
The ICC/ESOMAR International Code, revised June 2025, emphasizes accountability, transparency and human oversight as AI enters research. The accuracy literature on where that oversight holds up sits with AI open-end coding software for survey verbatims.
How Do You Turn Verbatims Into Something You Can Report?
Coding assigns a label to what a verbatim says; analysis is the separate step that decides what the pattern of labels means, and most open-end programs skip straight from one to the other. Plan the frame against the research question, sample before coding everything, and count the pattern before quoting it.
CDC's program-evaluation guidance on qualitative data frames the work as four steps in order, starting from the evaluation question the coding has to answer. Its advice is blunt: collect only as much data as the team can actually analyze, because collecting it is quick and analyzing it is not.
- Plan the frame against the research question, not the answers. Decide what the study needs to learn before a single verbatim is read, or the frame drifts toward whatever the coder finds interesting.
- Sample before coding everything. Read enough verbatims to see the vocabulary repeat, draft categories from that smaller pass, then apply the finished frame to the rest. A frame drafted first fits assumptions, not responses.
- Report a count beside every theme. A quote illustrates; the finding is how many verbatims landed in each code. Pull the two or three strongest quotes to show why, never instead of the count.
- Fix the volume before fielding, not after. The point budget above caps how many questions get asked; team capacity should cap how many get coded, with more than one coder on the same frame so interpretation stays a finding, not opinion.
Coding the finished frame against every verbatim is the mechanical step covered above; getting the frame right first is not, and no tool does it for you.
What Changes When Someone Can Answer Out Loud?
A typed open end and a spoken one are not the same instrument: the literacy requirement can leave out people with limited reading and writing skills.
Completion rates by survey mode. A Survey Research Methods experiment with 1,001 Spanish panelists found response lower when voice input was pushed, and text-only answers were the least complete of the four conditions tested. In a 29-country European Social Survey analysis, a model identified administration mode at 77.6 percent accuracy against a 54.5 percent chance level: modes are not directly comparable, and Pew's typed panel breaks off at about 1 percent.
A self-administered box keeps whatever the respondent typed first; an interview asks again. Four theory-based probes in a survey chatbot improved response quality, and 1,800 respondents assigned to probing, coding chatbots gave more detail at a cost to experience.
Alchemic's WhatsApp-native interviews take that shape: no link, no app, answers arrive by text or voice note in the thread already open. Alchemic publishes 57+ languages including Spanish, Arabic and Mandarin, with managed fieldwork or bring your own across 14 markets including the USA and the UK. Voice notes are the format the data arrives in for this population, and an AI-moderated interview probes them.
Where Open Ended Questions Are the Wrong Tool
An open end is the wrong instrument more often than a question bank admits. Three cases where something else works better:
- The answer belongs to a known list. Drug names, occupations, brands, countries. Web survey research comparing input fields found a lookup list produces far more machine-matchable answers than a text box, though it takes longer to complete.
- You need depth, not breadth. A self-administered box collects one unprobed paragraph; a moderated conversation collects the follow-up that beats 400 shallow verbatims.
- Nobody is funded to read the answers. Uncoded verbatims are a liability with a timestamp: decide who reads them, and how, before fielding the question. Qualitative research was never the cheap option, only the planned one.
Two or three well-aimed open ends beat a full box of questions, because somebody reads them.