TL;DR:
A leading question pushes the answer. Use neutral wording that would read the same to someone who loves the thing and someone who hates it.
Ask one question at a time and give a balanced scale with matching options on both sides.
Read each question aloud and ask what a person who disagrees would feel answering it.
You run a team and want to know whether the new weekly planning meeting helps. You write: "How much did you enjoy our helpful new planning meeting?" Every answer to that question is already tilted, because the question has told people what you hope to hear. Good survey questions are written so the person answering can say what they really think without feeling nudged.
Where leading questions come from
Bias in a question rarely looks like bias to the writer. It slips in through three doors: words that carry a judgment ("helpful", "wasteful", "obviously"), questions that pack two ideas into one, and answer options that lean to one side. Learn to spot those three and most problems disappear.
The checks to run on every question
Strip the adjectives: Take out any word that praises or blames. Ask about the thing, not your opinion of the thing. "The new meeting" instead of "the helpful new meeting".
Count the ideas: If the question contains "and" or "or" joining two things, split it. A person who likes one half and dislikes the other cannot answer honestly. This is the one question at a time rule.
Balance the scale: Offer the same number of positive and negative options around a neutral midpoint, and label them in matching words, such as Very dissatisfied, Dissatisfied, Neither, Satisfied, Very satisfied.
Cover every honest answer: Add "Not sure" or "Does not apply" where someone might truly have no answer, so they are not forced to invent one.
Test it on someone: Ask one person to answer aloud and explain how they chose. Where they hesitate, the wording is probably unclear or loaded.
Worked example: rewriting three biased questions
Here are three questions you might really see, each with a problem and a rewrite.
Original | What is wrong | Rewrite |
|---|---|---|
How much did you enjoy our helpful new planning meeting? | Loaded words ("helpful", "enjoy") assume the answer is positive. | How satisfied or dissatisfied are you with the new planning meeting? (5-point scale, Very dissatisfied to Very satisfied) |
Do you agree that the meeting is too long and starts too late? | Two ideas in one, and the "agree" form invites a yes. | Question 1: How do you feel about the length of the meeting? (Too short, About right, Too long). Question 2: How do you feel about its start time? (Too early, About right, Too late) |
How often do you use our excellent shared drive: daily, weekly or sometimes? | "Excellent" is a judgment, and the options overlap and skip "never". | In a typical week, how many days do you use the shared drive? (0, 1 to 2, 3 to 4, 5) |
Check the last rewrite as you would check a calculation: the ranges 0, 1 to 2, 3 to 4 and 5 cover every whole number of days from 0 to 5 with no overlap, so each person fits exactly one option. In the second rewrite, someone who thinks the meeting is the right length but starts too late can now say so; before, they could not.
Choosing scale labels
Pick one scale and keep it across the survey so people are not relearning it on every page. An odd number of points gives a true midpoint; an even number forces a lean. Neither is wrong, but decide on purpose. Label every point with words, not only the ends, so "3" means the same thing to everyone.
Keep direction consistent too. If 5 means "very satisfied" on one question, do not make 5 mean "very dissatisfied" on the next. Respondents move fast and will answer the wrong way without noticing.
Order, context and the answer people can give
Earlier questions colour later ones. If you ask "How stressful is your workload?" and then "How satisfied are you with your job?", the first question has already put stress in the respondent's mind. Put general questions before specific ones, and keep sensitive or emotional topics away from the questions you want unaffected.
Also check that every person can truthfully answer. A question about last month's meeting is unfair to someone who joined last week. Add a screening question or a "Does not apply" option, and tell people which period you mean, for example "in the last four weeks".
Mistakes to avoid
Starting with "Don't you think...", which signals the expected answer.
Asking about the future in a way that sounds like a promise: "Would you use a better dashboard?" Everyone says yes to better.
Putting your most important question after twenty tiring ones.
Using internal jargon the respondent may not share.
Changing the wording between rounds, which breaks comparison with last time.
Doing it faster
When you build the form, paste your questions into a document first and run the five checks on a printout. If you are drafting in a form tool, moving the answers into a spreadsheet is straightforward, and a neutral question gives you cleaner columns to count. DuckyHelper, which sees your screen and answers out loud when you press fn on a Mac, can read a draft with you and point at any question that sounds loaded, but the final wording is your call.
Frequently asked questions
Is it ever fine to ask a yes or no question?
Yes, when the thing is really binary, such as "Did you attend the session?" Avoid it when the true answer is a degree, because yes or no hides the middle.
How many questions should a survey have?
Only as many as you will act on. Each extra question costs you some completed responses, so cut any question whose answer would not change what you do.
Get started
