Scale what you learned from interviews — without corrupting the data. A badly worded survey doesn't measure reality; it manufactures the answer you accidentally asked for.
▸ Try the interactive toolSurveys are a quantitative discovery tool that test whether patterns found in interviews hold at scale. Where interviews give depth — one person's vivid account — surveys give breadth: how many users share the experience, how intensely, and whether it varies by segment.
The danger is that surveys feel objective while being extremely easy to bias. A leading question, a missing option, or a vague scale produces clean-looking numbers that mean nothing. Good survey design is mostly about not corrupting the data — and the cardinal rule is that a survey scales a pattern you already found, it doesn't discover one from scratch.
One idea per question · No leading language · Mutually exclusive, exhaustive options · Consistent, labelled scales · Neutral order (randomise where possible) · Pilot before you send
A survey's authority is also its trap: people trust a percentage more than a quote, even when the percentage came from a broken question.
| The shortcut | What it costs | What it gives you instead |
|---|---|---|
| Discovering with surveys | Using a survey to find a problem you haven't heard in interviews. | Surveys confirm and size known patterns — they're a poor discovery instrument on their own. |
| Leading questions | “How much do you love feature X?” bakes in the answer. | Neutral wording lets the real distribution show. |
| Double-barrelled questions | “Is it fast and easy?” — a yes/no can't answer both. | One idea per question keeps each answer interpretable. |
| Forcing false choices | Options that overlap or omit the real answer distort everything downstream. | Mutually exclusive, exhaustive options — plus an escape hatch — capture reality. |
Write the one thing this survey must answer. If you can't state it in a sentence, you'll write a sprawling survey that answers nothing well.
Often you need different surveys for different segments rather than one survey with branching everywhere. Decide who you're asking before you write a question.
One idea per question, neutral language, clean options, labelled scales. Read each question as a hostile respondent looking for a way to be misunderstood.
Send to a handful of people first. Watch where they hesitate, misread, or pick “other.” Every confusion in the pilot is corrupted data in the real run.
Collect, then slice by segment. An average across very different users often hides the real story that segmentation reveals.
A team wanted to know whether to build a requested feature and sent a survey asking “How valuable would feature X be to you?” on a 1–5 scale. The average came back high, so they built it. Usage was near zero.
The question had measured enthusiasm for a hypothetical, not real demand — a classic leading, hypothetical pairing. A better question would have asked about the last time they'd needed the underlying capability and what they did instead.
The artifact is the survey itself, but the discipline is the per-question checklist it must pass.
| Principle | Bad | Better |
|---|---|---|
| One idea | “Is it fast and reliable?” | Two separate questions |
| Neutral | “How much do you love…?” | “How would you rate…?” |
| Clean options | “0–5, 5–10, 10+” (overlap) | “0–4, 5–9, 10+” |
| Labelled scale | 1–5 with ends unlabelled | Each point or both ends labelled |
| Escape hatch | Forced choice | “Other / none of these” included |
Surveys don't discover — they confirm and size. Pointing one at an unknown problem produces confident numbers about a question you didn't understand well enough to ask.
The trap is that survey output looks like truth. A percentage carries more authority than a quote, so a biased survey does more damage than a biased interview — it launders a guess into a statistic. The defence is boring discipline: one idea per question, neutral wording, and a pilot before you ever hit send.
If you haven't heard it in interviews, a survey won't find it — it'll just measure your framing. Discover qualitatively, size quantitatively.
“How much do you love…” guarantees inflated results. Stay neutral.
“Fast and easy?” can't be answered cleanly. Split them.
Every ambiguity you didn't catch becomes noise in the real data. Always pilot.
Interviews find the pattern; the survey measures how widely it holds. Run in that order.
Opportunity Scoring is a specific survey design — importance vs. satisfaction — built on these same principles.
Survey-sized demand sharpens the Reach and Confidence inputs to RICE.
Advanced survey design, sampling, and analysis live in the discovery module.
Recall a survey you've taken — a product NPS, a feedback form. Find one question that breaks a principle: double-barrelled, leading, overlapping options, or an unlabelled scale.
Rewrite it to be clean. Notice how the fixed version would produce different, more honest data.
Almost every survey in the wild violates at least one principle. Spotting it is the same skill as writing one that doesn't.
Ready to go deeper · Advanced Survey Design →