Best practices

A good study comes down to two things: screening for the right people, and asking questions that get honest, useful answers. This page collects the practices that make the difference.

Some are rules worth following on every study, because they keep your responses genuine. Others are rules of thumb: good defaults that won't fit every case, called out as such. It's a solid self-check before you publish, by hand or after the setup chat drafts your study.

Screener

The screener has one job: it decides who gets in. These practices keep it doing that job.

• Mark every multiple-choice answer deliberately. A screener with unmarked answers just collects data. It doesn't screen. Mark each option Qualify/Disqualify (single-select) or Must select/May select/Disqualify (multi-select), and confirm none were left unset.
• Use Disqualify only for hard requirements, not "less ideal." Over-disqualifying is the costliest mistake: it slows recruitment to a crawl. Ask: would including this person invalidate a finding? If yes, disqualify. If they're just not your dream participant, let them through.
• Keep interview questions out of the screener. Anything beyond include/exclude dilutes the screener and slows recruitment. "What frustrates you about X?" is an interview question. The screener only needs "Do you use X?"
• Make options mutually exclusive and collectively exhaustive. No two answers may both be true at once (watch off-by-one ranges like "1 to 5" and "5 to 10"), and anyone who shows up needs an option that fits. Turn on "Other" if they might not.
• Aim for three to five questions. (Rule of thumb.) A long screener slows recruitment, and if it takes eight questions to qualify, the criteria are probably tangled. Find the one or two filters that actually decide inclusion and trust the rest to manual review. Deviating is fine when eligibility is genuinely complex.
• On a panel, add a trap option marked Disqualify that looks real. (Rule of thumb.) Without one, inattentive panelists click through and pollute your data. Make it a plausible-sounding fake.

Question guide

Coverage

• Cover every research objective with at least one question. If an objective has no question, the study can't answer it. Read down your objectives and, for each, name the question that produces data for it.

Question wording

• Don't lead or load. A leading question plants the answer it wants ("Don't you think…", "What problems do you have with…") and participants are usually too polite to push back. Strip the assumption and let them supply it. (A question inside a conditional section gated by a screener answer that already established the assumption isn't leading: the gate filtered to the audience for whom the framing is true.)
• Ask about what people did, not what they'd do. "Tell me about the last time you…" is reliable; "Would you use a feature that…" is a coin flip: people are bad at predicting their own behavior. (Reactions to something a participant is looking at right now, such as a concept, a prototype or a mockup, are fine; those are reactions to something concrete, not predictions.)
• Ask one thing per question. "What do you like and dislike about X?" is two questions, and you can't tell which half they answered. Split it, or ask one half and probe the other as a follow-up.
• Keep questions open-ended. If a participant can answer in one word, the question isn't doing any work. "Walk me through…" beats "Did you…"; "How did you decide…" beats "Was it easy…". (A closed question used as a checkpoint or filter is fine.)
• Don't ask the same thing twice. Duplicate questions waste interview time and add noise, not richness. Consolidate them or sharpen the difference. (Two questions on mutually-exclusive branches aren't duplicates: no participant sees both.)
• Match the question type to the data you want. Open questions give qualitative texture; structured types (survey questions, such as multiple choice, ranking and slider) give data you can compare across participants. Don't stack several structured questions in a row. Mix in open questions, so the guide does not feel like a form.

Structure & pacing

• Follow this order: warm-up → context → deep dive → exploration → wrap-up. (Rule of thumb.) One or two warm-up questions let participants settle before you ask them to recall specifics in detail.
• Five to seven sections is a comfortable range. (Rule of thumb.) One giant section feels monolithic; ten feels fragmented. The thing to watch isn't the count: it's one section that dwarfs the rest (say 80%+ of the questions), where fatigue and thin coverage creep in.
• Aim for roughly 15 to 30 minutes. (Rule of thumb.) Under 15 is usually too thin; past 45, participants fade and late answers get worse. The real question is whether each item earns its time, not whether the total clears some threshold.
• Put a short statement at each section transition. (Rule of thumb.) It signals a topic change and gives participants a beat to reorient. A guide that's only questions feels like an interrogation.

Follow-ups

• If a question has follow-ups on, write follow-up guidance. Without it, the moderator probes blind and you get generic "tell me more." Guidance tells it where to dig, e.g. "probe the specific pain points and the workarounds they use."
• Keep guidance on the current topic. Follow-ups are a depth tool, not a coverage tool. If the guidance reads like "also find out about pricing" on a discovery question, a real question got buried. Promote it to its own item.
• Calibrate the depth. Default to Low for most questions; reserve High for the one or two pillar questions per study. If the guide runs long, turn follow-ups down on questions that aren't adding nuance; if it runs short with good coverage, deepen them on the pillars; if it runs short with coverage holes, add questions instead.

Sensitive topics

• Frame sensitive topics neutrally, and don't avoid them. For AI, jobs, money, health: add a brief "there are no right or wrong answers, I'm just curious about your experience" statement before the section, and ask in a behavioral, non-judgmental form.

Intro & close

• Keep the intro to project context in two friendly sentences. Don't mention recording (already consented to; raising it is unsettling), time estimates (usually wrong; they make people rush), or future contact (don't promise what the researcher might not deliver).
• End with a simple thank-you. No "we'll be in touch," no five ways of asking "anything else?", which bleeds the interview into "actually, one more thing…" Thank them and end clean.

Media & tasks

• Keep questions with the media, not before or after. The visual stays on screen while the moderator asks about it. Questions placed next to the media instead of with it fire after it disappears, and the participant is answering about something they can no longer see.
• Use real URLs only. No example.com, no placeholder links, no "this is roughly the link." If you don't have the real URL yet, leave the media empty and add it later: fake URLs ship to participants.
• Make tasks prescriptive, granular, and nested. Prescriptive: no ambiguity about what to do. For usability tests, a guiding statement orients the participant (where to be, what to look at). Granular: several short tasks beat one long one. Nested: a task sits under the image, video, or prototype it's about.
• In front of a prototype, use a task, not a question. Open questions about a static prototype get shallow answers: the participant is describing a screenshot. A task ("find where you'd update your billing info") forces interaction; for open exploration, frame that as a task too ("explore and tell me what stands out"). Any question comes after the task.

Choosing the right element

• Use a slider for ratings, scales, and magnitudes. Do not render a 1 to 10 satisfaction question as ten radio buttons. Sliders show the answer is on a continuum and produce cleaner, comparable data. Use multiple choice only when the options are categorical and unordered.
• Mix element types. (Rule of thumb.) A guide built entirely from open questions misses structured signal. Mix in a statement at section transitions, a multiple-choice item or two for quick quantitative reads, and a ranking when the topic is about prioritization. Not every guide needs all four, but a guide with only questions is usually under-specified.

Related