A customer can like the person delivering a service and struggle with the document that person sends.
Ask whether the service is “helpful and easy to use” and the answer has two jobs. A positive response cannot tell you which job succeeded. A negative response leaves the repair equally unclear.
Start with the decision. Give each question one piece of evidence to collect. Keep the respondent's task visible through the wording, answer choices, and routing.
The questionnaire below belongs to a fictional document-delivery service. Its questions and review worksheets are proposed working methods. They are not a validated research instrument.
Define the Decision Before Writing Questions
The fictional team has one decision: whether to test a revised contents page for its monthly delivery pack.
It wants to learn how customers locate the current checklist. It has not decided that a redesign is necessary. A finding that people find the checklist easily would leave the proposal without its assumed problem.
Eligible respondents are people who personally tried to locate that checklist in the latest pack. Account sponsors who never opened it cannot report that experience. They may have useful purchasing views, but those views belong to a different question.
Write a decision card before drafting:
| Field | Fictional entry |
|---|---|
| Decision | Should the team test a revised contents page? |
| Task under study | Find the current checklist in the monthly delivery pack |
| Reference event | The respondent's most recent attempt in October 2026 |
| Eligible perspective | A person who personally attempted that task |
| Needed evidence | Reported difficulty, task outcome, and where the search began |
| Outside scope | Satisfaction with the account team, prices, and future purchases |
Keep separate fields for difficulty and outcome. Someone might eventually find the checklist after a difficult search. Completion alone would hide that distinction.
Before commissioning a survey, AAPOR recommends checking whether another method better answers the research question. Here, the survey would describe reported experiences. Observing a task would be a separate way to investigate what happened.
If you still need customers' language for the task, begin with customer interviews for B2B content research. Do not ask a standardized question whose central term nobody has defined.
Repair the Question Before Polishing the Form
Put each draft beside the concept it is meant to measure. If you cannot name one concept, split or remove the question.
Pew Research Center's question-writing guidance addresses wording, separate concepts, answer categories, and question order. Apply those checks before deciding how attractive the form should look.
These fictional repairs change the measurement, not merely the tone:
| Defect | Draft question | Proposed revision |
|---|---|---|
| Leading premise | How helpful is our improved contents page? | On your most recent attempt, how easy or difficult was finding the current checklist? |
| Two concepts | Was the pack easy to find and understand? | Ask about finding the checklist. Ask about understanding its instructions separately, if that evidence is needed. |
| Unclear period | Do you usually find the checklist quickly? | Think about your most recent attempt to find the current checklist in the October 2026 pack. |
| Overlapping choices | How many attempts: 0–2, 2–4, or 4–6? | If attempt count is needed: 0, 1, 2, 3, 4 or more, or cannot recall. |
The first revision removes “improved.” The team has proposed a change; the respondent has not established that the change helps.
The second repair exposes a scope decision. Understanding instructions may matter, but adding it requires another question and another use for the answer.
The final repair still needs a definition of “attempt.” Does opening the pack twice count as two attempts? Drop the count unless the team needs it and respondents can answer consistently.
Do not preserve a question because someone already designed a chart for it.
Give the Answer Choices Clear Boundaries
For the fictional difficulty item, use one ordered scale: very easy, somewhat easy, neither easy nor difficult, somewhat difficult, very difficult, and cannot recall.
“Neither” records a position on the scale. “Cannot recall” records unavailable recall. Neither means the same thing as leaving the item blank.
Keep that distinction in the export. Do not turn every nonpositive response into dissatisfaction. An unanswered item supplies no rating to classify.
Use task eligibility to handle “not applicable.” A person who never attempted the task exits before the difficulty item. A person who attempted but abandoned it remains eligible to describe the difficulty.
For a question about where the search began, offer mutually exclusive starting locations and an “other” choice. Ask for the first location, not every location visited. That keeps one answer tied to one event.
Review the proposed locations with people who recognize the pack. If they consider “contents page” and “opening page” the same place, merge or clarify the labels before launch.
This is a design choice for this questionnaire. It is not a universal scale or a benchmark that can be compared with unrelated satisfaction measures.
A Short Fictional Questionnaire With Routing
Introduce the task before showing the first item:
Think about your most recent attempt to find the current checklist in the October 2026 monthly delivery pack. Answer from your own experience. You may skip any question. Do not enter names, account details, file links, or information from customer records.
Q1. During October 2026, did you personally try to find the current checklist in that month's delivery pack?
- Yes.
- No.
- I am not sure which pack this means.
- Prefer not to answer.
If Q1 is yes, show Q2 through Q5. For every other response, end with a neutral thank-you. If Q1 is blank, offer the same exit. Keep those four outcomes separate in the export.
Q2. On that most recent attempt, did you find the current checklist?
- Yes.
- No.
- Cannot recall.
- Prefer not to answer.
Show Q3 to everyone routed from Q1 yes, including people who did not find it. Do not skip the difficulty question because the task failed.
Q3. On that attempt, how easy or difficult was finding the current checklist?
- Very easy.
- Somewhat easy.
- Neither easy nor difficult.
- Somewhat difficult.
- Very difficult.
- Cannot recall.
- Prefer not to answer.
Q4. Where did you first look for the current checklist on that attempt?
- The pack's contents page.
- A section of the pack after the contents page.
- The message containing the pack.
- Somewhere else.
- Cannot recall.
- Prefer not to answer.
Q5. What, if anything, made finding the checklist difficult on that attempt?
Provide an optional text box. Repeat the instruction to exclude names, links, and customer records. Someone who found it easily can write “nothing” or skip the item.
Q5 asks for context. It does not ask for a redesign suggestion or a commitment to buy something. A comment can guide further investigation without becoming the specification for a new feature.
Test the route where Q2 is blank but Q3 is answered. The fictional design permits that sequence. The eventual report must preserve each item's denominator instead of discarding the entire response automatically.
Pretest Comprehension and Answerability
Run a pretest with people resembling the intended respondents. AAPOR describes pretesting as a way to examine interpretation and answer formation.
For this proposed pretest, ask participants to complete the draft in the intended format. Then ask them to explain what “current checklist” meant. Ask which event they recalled and how they selected an answer.
Avoid teaching the intended meaning before they answer. If the facilitator supplies the definition, the session no longer shows whether the question communicates it.
Include different eligible experiences in the pretest: finding the checklist, abandoning the search, and being uncertain about the pack. These are scenarios to examine, not a prescribed sample size.
Separate two defects in the notes. A comprehension problem means the person interpreted the item differently. An answerability problem means they understood it but lacked the requested knowledge or recall.
Inspect the form on a small screen. Confirm that long options remain readable and that optional items remain optional. Follow every exit path. Review the exported field values after each path.
Do not treat a functioning form as proof that the questions measure the intended task. Keep the wording review and the routing check as separate results.
Keep a Question-Change Record
Save the tested version before revising it. Give each item a stable identifier, such as Q3, even when its wording changes.
Use this fictional record to explain the revision:
| Item | Pretest observation | Change | Reason | Recheck |
|---|---|---|---|---|
| Q1 | Participant interprets “latest” as a prior month's pack | Name October 2026 | Fix the reference event | Ask which pack the revised item names |
| Q3 | Participant rates the account team | Repeat “finding the current checklist” | Return to task difficulty | Ask what the rating describes |
| Q4 | Participant cannot distinguish two location labels | Replace ambiguous labels with pack wording | Separate starting locations | Ask where each option points |
| Routing | A no response still sees Q3 | End after Q1 no | Keep ineligible ratings out | Export a synthetic no-response path |
Record the editor, version, and review date alongside the table. Keep the record free of participant identities and private examples.
Before launch, agree which version is final. If a defect forces an in-field change, preserve the versions and response periods separately. A later analyst needs to know which question each person saw.
For repeated surveys, Pew advises attention to wording and questionnaire context when measuring change. A revision belongs in the comparison notes rather than disappearing from the record.
Keep the Research Invitation Separate From Selling
A research invitation should explain the task, who should answer, and how participation works. Route urgent service requests to the ordinary support channel. The survey is not an incident response queue.
For the fictional pack study, invite personal experience with one task. Do not attach a renewal offer, product pitch, or account review request. AAPOR distinguishes research from selling and predetermined outcomes.
Describe handling accurately. If an invitation token can connect an answer to an account, do not call the survey anonymous. Decide who can access that mapping before collecting responses.
Keep commercial follow-up outside this questionnaire. The proposed research has no field for contact details, contract values, or uploaded documents. Its optional comment box explicitly excludes customer records.
If existing support-ticket summaries already answer the narrow question, review that evidence before asking customers to repeat it. Use authorized summaries rather than copying private conversations into the form.
Keep search planning downstream of the evidence. B2B keyword research can explore how people search for a verified task. It cannot repair a survey question that asked three things at once.
Write the question that the decision needs. Test what the respondent hears.
Sources
- Pew Research Center: Writing Survey Questions. Question wording, concepts, answer categories, and comparability. Checked 2026.10.28.
- AAPOR: Best Practices for Survey Research. Research objectives, pretesting, respondent choices, and separation from selling. Checked 2026.10.28.