How to use the Survey Question Builder

Paste any survey question. The tool checks for: leading words (amazing, excellent — biases positive), double-barreled construction (asks two things), corporate jargon (confuses non-expert respondents), and length (over 120 chars reduces completion). When bias is found, you get an auto-rewrite with leading words stripped.

Why this tool matters

Bad survey questions produce bad data. The biggest culprits are leading words (loading the respondent toward a positive answer), double-barreled construction (no clean answer possible), and jargon (respondents skip what they cant understand). A 30-second bias check before sending a survey improves data quality by 20-40%.

Common use cases

• Pre-launch QA on customer satisfaction surveys
• Reviewing NPS follow-up questions
• Auditing post-purchase email surveys
• Refining product feedback forms
• Helping research teams produce cleaner questions
• Training new researchers on bias-free question writing

The most common bias patterns

Leading: How much do you love our amazing product? — biases positive. Double-barreled: How satisfied are you with our product AND support? — asks two things. Jargon: How effectively does our platform leverage your workflow? — confuses respondents. Each pattern systematically distorts the data you collect.

Frequently Asked Questions

Should I use 5-point or 7-point Likert scales?
7-point gives more granularity but also more “neutral” answers. 5-point is faster to complete and works for most cases. NPS uses 11-point (0-10) specifically because it captures detractor distinction at low scores.

Whats the right number of questions per survey?
Under 10 for completion rates above 80%. 10-20 acceptable for engaged audiences. Over 20 drops completion below 50%.

Should I ask demographics first or last?
Last. Demographics feel intrusive — putting them first triggers dropoffs. Save them for the end where respondents have invested time and want to finish.

How do I deal with “no opinion” responses?
Allow “Not applicable” or “Dont know” — forcing an answer produces fake data. Just track the no-opinion rate separately.