Every researcher knows what a leading question looks like in a textbook example: "Don't you agree that our excellent customer service exceeds your expectations?" Nobody writes that. The wording problems that actually make it into fielded studies are quieter, and they survive review because the person who wrote them already believes the premise.
1. Leading language that isn't obviously loaded
Compare: "How much has our new dashboard improved your workflow?" versus "How, if at all, has our new dashboard changed your workflow?" The first assumes improvement occurred and asks only for a magnitude. A respondent who thinks the dashboard made things worse has no honest way to answer it. This pattern shows up constantly in post-launch satisfaction surveys, where the business already has a stake in the answer.
The tell: if removing the adjective ("new," "improved," "innovative") from your own question makes it feel incomplete, the adjective was doing the work of the question.
2. Double-barreled questions
"How satisfied are you with our product's price and quality?" asks two questions and records one number. A respondent who loves the quality and hates the price has no correct answer, so they'll pick whichever matters more to them that day, and you'll never know which. This is one of the most common defects found in questionnaire audits, precisely because each half of the sentence sounds reasonable on its own.
The fix is mechanical: if a question contains "and," check whether it's joining two nouns that could each get a different rating. If so, split it into two questions.
3. Unbalanced response scales
A scale of Excellent / Very Good / Good / Fair / Poor has three positive options and one negative one. Respondents cluster toward the middle of whatever scale you give them, so an unbalanced scale mechanically inflates the positive skew regardless of true sentiment. This is a distinct failure from leading language: the question text can be perfectly neutral and the scale will still distort the result.
A balanced scale needs equal positive and negative options around a true midpoint: Excellent / Good / Neutral / Poor / Very Poor, for example.
4. Order effects inside a single question's options
Presenting options in a fixed order (rather than randomized) consistently favors whichever options appear first in a list, an effect that's well documented across self-administered surveys. If you're asking respondents to choose among more than three or four options, and the order isn't randomized, your ranked results are partly measuring position, not preference.
The one-question audit
Before fielding anything, read each question as if you already believed the opposite of your hypothesis. If the question feels awkward or impossible to answer honestly from that position, it's leading. This catches the majority of wording defects in about the time it takes to read the questionnaire once, straight through.