Research ·
Better surveys start with better questions, not more questions
A practical approach to bias, sampling, reliability and decision-ready evidence.
By Harry Savigar

Better surveys start with better questions, not more questions. A practical approach to bias, sampling, reliability and decision-ready evidence. This article is part 3 of a five-part exploration of what makes Asketta useful beyond a conventional form.
What makes a survey question unreliable?
Leading wording, double-barrelled questions and unclear timeframes can all distort answers. 'How excellent was our service?' assumes a positive experience. 'How satisfied are you with delivery speed and quality?' asks two questions at once. Better versions use balanced options, one idea per question and a defined reference period.
How can an AI-assisted designer help?
Start by describing the decision the survey must inform, the population you want to hear from and what you already know. An assisted design process can propose questions, flag obvious ambiguity and suggest alternative response formats. Human review remains necessary, particularly where wording could disadvantage groups or shape a commercial conclusion.
Turn this article into a conversation
Tell Harry what makes your surveys difficult to trust and what would make the next article more useful. Answers can be stored with your consent; a contact email is optional.
Participation is optional. Submitting answers requires your consent; individual answers are not public. Learn how Asketta works on websites.
How much sample size is enough?
There is no universal number. The right sample depends on the population, desired precision, expected variation, recruitment method and whether the goal is to compare groups. A large convenience sample may still be unrepresentative. Before collecting data, document who is eligible, how participants will be selected and how uncertainty will be reported.
What does reliability actually mean?
Reliable measurements produce reasonably consistent results under comparable conditions. For a multi-item scale, check whether questions measure the same concept and whether repeated measurement is stable when change is not expected. Reliability is not the same as validity: a consistent survey can still measure the wrong thing.
How do you turn responses into defensible findings?
Review missing answers, subgroup sizes, wording effects and possible non-response bias. Separate observed findings from interpretations. Report counts alongside percentages; avoid dramatic claims from tiny samples. If comparing teams or time periods, check whether question wording and sampling remained consistent.
Where does Asketta fit?
Asketta can support questionnaire creation, adaptive follow-up and structured response analysis. More advanced evidence checks—such as explicit sampling assumptions, bias review and reliability assessment—should be treated as an analyst-supervised workflow and validated before any research-grade promise is made.
What should you do next?
Start with a specific decision and a short set of questions. Explore Asketta at https://asketta.ai and test a focused questionnaire before scaling. The most useful interaction is one that leads to a clearer next step.
Related perspective: https://asketta.ai/blog/stop-broadcasting-start-a-conversation
Frequently asked questions
- Can AI remove survey bias completely?
- No. AI can flag risks, but study design, sampling and human judgement remain essential.
- Is a high response count the same as representative research?
- No. Representativeness depends on how participants were recruited and who did not respond.