A spoken method for drafting survey questions tied to one decision.

Dictating a Prompt for survey question design works best when the survey is tied to one decision. Before you speak, name the decision the results will inform, the person who will make it, and the date it is needed. A survey that exists to satisfy curiosity produces questions that are interesting but not actionable. Say the decision first in your prompt, then ask for questions that could change it. If the answer cannot change anything, the question does not belong. This single test removes more weak questions than any style guide, and it is easy to state in one spoken sentence before the drafting begins.
Write the research question in the language of the decision, not the language of the survey. For example, instead of asking about satisfaction, ask whether the team should keep, change, or stop a weekly process. Say who currently owns the process and what would count as evidence for each option. The assistant can then propose questions whose answers map to those options. If a proposed question produces an answer that supports every option equally, it is not useful. Speak the mapping explicitly and ask for it in the output, so you can see how each question earns its place. A single decision also keeps the survey short enough for a busy audience.
Set the scope with numbers you can defend. Say the maximum number of questions, the expected completion time, and whether the survey is one page or several. A short survey gets finished; a long survey gets abandoned halfway, and the answers you lose are rarely random. Ask for the estimated time beside each question so you can cut the expensive ones. If the survey must cover several topics, decide which topic gets the most space before you speak. Scope is a budget, and a spoken prompt should state the budget up front rather than discovering it after the draft arrives.
Name the audience and the recruitment path. Say whether respondents are customers, employees, students, or applicants, and how they will be invited. A survey sent to a broad list needs more screening questions than one sent to a known group. Mention the context in which people will answer: on a phone between meetings, at the end of a service call, or in a classroom. Context shapes question length and wording. A question that works on a desktop form may fail on a small screen. Speak the context and ask the assistant to keep each question short enough for it. Say whether the invitation explains the purpose, since that affects whether people start.
Describe the respondent as a person, not a segment label. Say what they already know about the topic, what vocabulary they use, and what they might misunderstand. If respondents are new to the process, avoid internal terms and spell out acronyms. If they are experts, skip the basics and ask for precision. This description lets the assistant choose wording at the right level instead of writing for a generic reader. Speak one example of a typical respondent and one example of an unusual one, so the questions work for both without becoming vague or overloaded with explanation. Keep that description in the template so every survey for this audience starts from the same picture.
Say what respondents can and cannot answer. People can report their own experience, their recent actions, and their preferences. They cannot reliably report someone else's behavior or explain the cause of a problem they only observe. Ask the assistant to phrase questions around what the respondent actually knows. If you need information about a group, ask about the respondent's own contact with that group rather than asking them to speak for it. This rule prevents confident answers to questions that should never have been asked of that audience. When a question needs an estimate, ask for a range rather than a precise figure.
Choose the response format for each kind of question. A yes-or-no question suits a clear decision; a scale suits a degree; a short list suits a recall task; an open question suits a reason. Say which formats are allowed and how many of each you want. For scales, specify the endpoints in words, such as not at all and every day, rather than numbers alone. For lists, say how many options and whether an other option is allowed. Format is part of the question, and a spoken brief should decide it before drafting begins. Ask for the format beside each question so the draft can be checked at a glance.
Plan for the respondent who wants to stop. Ask for questions that are easy to skip and a structure that does not punish a partial response. Say whether every question is required and, if not, which ones can be left blank. A survey that forces an answer to a question the respondent cannot address produces noise that looks like signal. Ask the assistant to mark optional questions and to keep the most important ones early. Respect for the respondent's time is not a courtesy; it is a condition for usable results, and it should be written into the prompt. A short survey that people finish is worth more than a thorough one they abandon.
Speak the rules as a checklist the assistant can follow. Start with one idea per question, plain wording, no double negatives, and no leading phrases. Add the rules specific to your topic: do not assume a behavior, do not ask two things at once, and do not offer an answer the respondent cannot choose. Say the order, from broad to specific. Then ask the assistant to show the checklist beside each drafted question. A visible rule list makes the draft easier to audit and easier to revise when a question fails. Number the rules so you can refer to one during review without repeating it.
Ban the constructs that quietly distort answers. Leading questions suggest the desired answer; loaded questions combine a fact and a judgment; double-barreled questions ask about two things and record one answer. Say these bans in plain language and give one example of each in your own topic. Ask the assistant to label any question that comes close to a ban instead of silently rewriting it. Labels let you decide whether the question is acceptable in context. A short banned-construct list is more effective than a long style guide read once and forgotten. Labeling is useful even when you keep the question, because it records the risk.
Specify the answer options before the questions. For each closed question, say the options, whether they are mutually exclusive, and whether an other or none option is allowed. Options shape the question as much as the stem does, and a missing option forces respondents into a wrong answer. Ask the assistant to check that every option is plausible for the audience and that no two options overlap. If the options cannot be settled, the question is not ready. Speak the option rules once and apply them to every closed question in the set. Write the option sets down before the stems so the wording cannot drift.
Decide how to handle sensitive or difficult topics. Say whether the survey asks about money, health, work performance, or conflict, and whether respondents can answer anonymously. Ask for neutral wording that does not imply judgment, and for an explicit option to decline. A question that traps the respondent between a painful admission and a false answer will produce neither truth nor completion. If the topic is too sensitive for the channel, say so and remove it rather than softening the question until it means nothing. State the anonymity rule in the invitation and keep it consistent with the question design.
Read the draft as a respondent would, one question at a time. For each question, say the answer you would give and check whether the options cover it. Then ask whether the question could be read two ways; if it can, rewrite it. Check that the order does not create expectations, such as asking about a problem before asking whether the problem exists. Look for questions that repeat an earlier one in different language. An audit is not a proofread; it is a test of whether each question can produce a usable answer. Read the questions in the order they will appear, not the order they were drafted.
Check the draft against the decision from section one. For each question, name the decision it informs and the option its answer would support. If you cannot name one, cut the question. This test often removes a third of a first draft and shortens the survey without losing anything the decision needs. Ask the assistant to include the mapping in the output so the audit is quick. When a question is interesting but not decision-relevant, move it to a separate list for future research rather than keeping it in this survey. Keep the mapping in the appendix so the decision trail stays visible.
Test the wording with two people who resemble the audience. Ask them to read each question aloud and explain what they think it asks. Where their explanation differs from yours, the question is ambiguous. Do not defend the wording; rewrite it. Note the phrases that caused confusion and add them to the prompt as banned constructions. Two readers catch most problems, and the exercise takes less time than analyzing incomplete responses later. Wording problems are cheapest to fix before the survey is fielded. Record the exact phrase that caused confusion, not just the question number, so the rewrite is easy to explain.
Keep a version log for the question set. Record the date, the change, the reason, and the decision each version served. When a stakeholder asks why a question disappeared, the log answers in one line. When a later survey covers a similar topic, the log shows which questions worked and which confused respondents. Use the app to dictate audit notes beside the draft, so the reasoning stays with the questions. A version log turns a one-time design into a reusable method for the next survey. The log is also the fastest way to train a new analyst on the survey conventions.
Send the survey to a small group that resembles the real audience and watch how they respond. Ask them to note the question where they hesitated, the option they wanted but did not find, and the point where they considered stopping. Do not ask whether they liked the survey; ask what they did. A pilot is a test of the instrument, not a preview of the results. Fix the questions before the full launch, and keep the pilot responses out of the final analysis so the sample stays comparable. A pilot of five people is enough to expose wording problems before a wide send.
Measure completion and drop-off by question. If many respondents stop at the same point, the problem is probably the question before it or the length of the section. If a question is often skipped, it may be optional for good reason or badly worded. Review both possibilities before cutting it. Say these checks in your prompt when you ask for the next version, so the assistant proposes a shorter path. A survey that respects attention gets more complete answers than one that merely asks for them. Record the drop-off point for each respondent so the pattern is visible without rereading every response.
Review open-ended answers for ambiguity, not for agreement. Pilot answers show whether respondents understood the question, whether they answered in the expected format, and whether the options missed a common case. If several answers fall outside the options, add an option or convert the question to an open one. If answers are consistently off-topic, the question is unclear. Use these findings to revise the prompt rules, then run a second small pilot if the changes are substantial. Keep the pilot answers in a separate note from the final results, and compare the two only after the fixes. That comparison is quick.
Document the final instrument and the prompt that produced it. Keep the question set, the rules, the audit notes, and the version log together. Anyone who wants to repeat this workflow can download Voiskey, speak the decision, audience, scope, and rules into the drafting window, and produce a first set of questions in minutes. The tool speeds up the drafting; the audit and pilot still decide whether the survey is ready. That division of labor is what keeps a fast draft from becoming a fast mistake. A short handover note saves the next team from repeating the same design debate.
Voiskey is an AI voice typing app that works in the drafting window you already use. Press a shortcut and speak the decision, audience, scope, and question rules; the prompt appears at the insertion point. Filler words, stutters, and self-corrections are removed, so the brief arrives as a clean checklist the assistant can follow.
Voiskey learns names, jargon, acronyms, and spelling preferences. Say respondent segments, program names, and scale labels once, and it keeps your spelling in later prompts. Check the drafted questions for drift, then update the brief before sharing it, because inconsistent labels across questions confuse respondents.
Yes. Voiskey removes filler words, stutters, false starts, self-corrections, and grammar slips as you speak, and it adapts punctuation and formatting to the task. A spoken brief with the decision, rules, and banned constructs arrives as a clean set of instructions, which makes the drafted questions easier to audit.
Voiskey supports more than one hundred languages and translation. You can dictate the design brief in one language and ask for questions in another, then review each translated question with a fluent speaker to check meaning and tone. Translation helps reach respondents; a fluent reviewer still decides whether a question reads naturally.

Download Voiskey for Mac and Windows and dictate your next survey brief.
Use it across macOS, Windows, iOS, and Android