12 leading question examples—and better survey rewrites

What makes a survey question leading?

A useful first check is to ask whether the wording contains the answer the team hopes to hear. Praise, social pressure and assumed benefits can all narrow the respondent’s task. The issue is the pressure built into the instrument, not whether the topic itself is positive or negative.

Pew Research Center’s methods guidance explains that wording and response options can affect answers. The rewrites below are MTB’s own teaching examples. They are suggestions to adapt and pretest, not evidence that a particular rewrite will eliminate every source of bias.

Sources: Pew guidance on question wording

12 before-and-after examples

These examples use fictional products and situations. Some questions have more than one flaw. Removing leading language may still leave an eligibility problem, an unclear time period or two judgments in one sentence.

A three-pass rewrite method

Pass one: underline the assumptions. Look for adjectives such as excellent, effortless and improved. Look for a presumed action: why did you buy, when will you switch, or which benefit convinced you. Keep necessary factual context, but remove promotional framing from the measurement question.

Pass two: recover the intended measurement. Ask the researcher what decision the answer should inform. If the real concern is whether readers know the total price, a favorable-value rating will not resolve it. Ask for the price they understood, then compare that answer with the actual offer.

Pass three: test the exit routes. Can someone disagree, have no preference, lack experience, or say the information is insufficient? Provide those routes where appropriate. “None” and “cannot judge” should not be interchangeable.

Finally, read the question with its answer choices and the preceding screen. A neutral sentence can still follow a promotional paragraph that teaches the desired answer. Our question-order guide covers that separate problem.

Leading, loaded and double-barreled are different checks

A leading question steers toward an answer. A loaded question builds in a disputed or unverified assumption. A double-barreled question asks for a single response to separate judgments. These problems can occur together, but the repair depends on the actual issue.

Consider “How much do you love our cheap and reliable service?” Removing love leaves cheap and reliable, which can receive different answers. Asking two questions may be more useful than polishing the original sentence.

An occurrence of “and” is only a review signal. “Which research and development team do you work with?” can name one team. A mechanical checker cannot decide that every conjunction makes a question invalid.

Check that the rewrite asks the same question

Keep the original beside the revision. Name the judgment you want to measure, then check whether it survives the rewrite. “What does this offer include?” measures understanding; “How appealing is this offer?” asks for a preference.

Remove an assumed experience or a suggested answer without adding product facts. If the question combines two judgments, split it and make the answer choices fit each part. Keep an uncertainty option when the respondent could reasonably lack enough information.

Before you send the survey

Have relevant people explain what they think each question is asking. Ask them how they chose their answer, particularly where wording has changed. Record revisions so the final questionnaire can be reproduced. The Census instrument-development standard treats pretesting as part of questionnaire development; this article’s checklist is a practical application, not an official instrument.

Sources: Census guidance on instrument development and pretesting

• Check one intended judgment per question.
• Read the complete answer list, not just the question.
• Keep criticism and uncertainty possible.
• Review context and question order.
• Retain the original and revised wording in the study record.

Common questions

Are all yes/no questions leading? No. “Did you use the search box?” can be a straightforward behavior question when the task and time period are clear. An open question such as “Why is our product better?” can be strongly leading.

Can software prove a question is unbiased? No. Automated flags can help reviewers find problems, but absence of a flag is not validation. Context, audience interpretation, sampling and administration still matter.

Should you remove every adjective? No. Some adjectives define the object being studied. The question is whether the wording provides necessary information or quietly tells the respondent how to judge it.

Repair the question and its answer choices together

Suppose the decision is whether the total price needs a clearer position on a checkout page. The draft “How much do you love our easy checkout?” with Good, Very good and Excellent as its only answers cannot answer that decision. Changing love to rate would still leave three favorable options and no specific task.

A usable revision starts after a task to find the total price. Ask the difficulty question below, then ask for the price the person found and what, if anything, made it hard to find. Record the correct total separately; do not reveal it before the answer. This changes the measure from general affection to a task-specific judgment, so document that choice rather than claiming the two versions measure the same thing.

Sources: Run the original question in the checker; Run the revised questionnaire; Edit and download the complete checkout template

What the checker found in this example

Running these exact inputs through rules version 1.0 on September 23, 2026 produced three signals on the original: a leading assumption, scale balance, and possible need for an uncertainty or no-experience option. The three revised question blocks produced no signals. These are recorded software outputs on constructed text, not responses from participants.

The tool did not decide whether the question matched the business decision, whether the price was visible, or whether the respondent found the correct amount. Those checks still need the study context and actual responses. Fewer signals show that listed wording patterns were removed; they do not establish less bias or a better survey.

Sources and scope

1. Writing Survey QuestionsPew Research Center

   Methods guidance on wording, response options and order. The examples and review workflow here are original MTB applications.

2. Statistical Quality Standard A2U.S. Census Bureau

   Primary standards for developing and pretesting data-collection instruments; not certification of these templates.