Write participant screener

Write a Participant Screener

Turn a participant profile into a screener that finds real target users, filters out professional participants, and never reveals what the study is about.

When to use / when NOT to use

Use when: a research plan or test plan names a participant profile and N, and you now need to actually find those people — screener questions, recruiting channel, incentive, schedule.

Do NOT use for:

• Deciding WHO the participants should be or HOW MANY — write-research-plan (general studies) and plan-usability-test (usability tests) own profile + N. If no profile exists, go there first; a screener without a profile is guessing.
• Writing a survey that collects research data — write-survey. A screener is a gate, not an instrument: screener answers decide who gets in; they are never analyzed as findings.
• The questions asked INSIDE the session — write-interview-guide / write-task-scenarios.

The core principle (NN/g)

A screener must elicit specific information about the respondent while concealing the study's purpose. Every question is a tension between those two goals. If a respondent can tell what answer gets them the incentive, some will give that answer — and your study runs on impostors. Every rule below serves one of the two goals.

The method

1. Pull the profile. From the research plan: target behaviors, exclusions, quota splits, N. Copy them verbatim — do not reinterpret.
2. Translate every criterion into behavior. Demographics are proxies; behaviors are the thing itself (conversion table below). Keep a demographic criterion only when the study genuinely requires it (e.g. a quota for assistive-tech users, a legal age floor).
3. Draft questions using the type tree (below). One criterion per question — never two.
4. Mask intent on every question: lift each question one level of abstraction above the study topic and surround the target option with plausible distractors (rules below).
5. Order exclusion-first (ordering section below): cheapest, most-exclusionary knockouts at the top.
6. Write accept/reject/quota logic per question. Every question carries explicit logic: TERMINATE if…, MUST include…, QUOTA: n per cell. A question with no logic attached is decoration — cut it.
7. Add the professional-participant traps (red-flag section below).
8. Plan the recruiting mechanics: channel, incentive, over-recruit buffer, scheduling (logistics section below).
9. Pilot on 2–3 people who should obviously pass and one who should obviously fail. If the fail-case passes, or a passer can name the study topic, revise before launch.

Behavioral, not demographic

The question is never "who are they" — it's "what have they done, how recently, how often."

Demographic proxy (don't)	Behavioral criterion (do)
"Age 25–40, tech-savvy"	"Completed an online booking or purchase on their phone in the past month"
"Small-business owners"	"Personally approves invoices or payroll at a company of <50 people"
"Fitness enthusiasts"	"Tracked a workout with any app or device 3+ times in the past 2 weeks"
"Frequent travelers"	"Booked 3+ trips with an overnight stay in the past 12 months"
"Our target persona: busy parent"	"Schedules appointments for at least one other household member"

Recency windows matter: "have you ever" recruits people working from stale memory. Default to the window the study needs — usually 1–3 months for routine behavior, 12 months for rare events.

Question type tree

What does this criterion need?
├─ Verify a target behavior happened
│    → Multiple-choice, SELECT ALL THAT APPLY, target hidden among distractors
├─ Verify frequency/recency of that behavior
│    → Single-select with graduated bands; accept band defined in logic
├─ Split accepted people into quota cells (segment, channel, device)
│    → Single-select, mutually exclusive + exhaustive options
├─ Check articulateness (interviews & think-aloud studies only)
│    → One open-ended question: "Walk me through the last time you…"
│      (2+ specific sentences = pass; blank, one word, or generic = reject)
└─ Logistics (availability, device, recording consent)
     → Closed questions, LAST in the screener

Never yes/no on the target behavior ("Do you book salon appointments online?") — it telegraphs the desired answer. NN/g's canonical form: ask "what activities do you do online?" — not "do you play online games?"

Distractor rules (masking done right)

• Same specificity as the target: if the target is "scheduled a personal-care appointment online," a distractor is "ordered groceries online" — not "used the internet."
• Same plausibility and social value. No option should feel like the impressive answer or the throwaway.
• 4–6 distractors per select-all question; more hides the target better but tires respondents.
• Always include "None of the above." Without it, respondents are forced to pick something and you can't detect non-qualifiers.
• The screener's public title masks too. "Short survey about everyday scheduling habits" — never "Booking-app study." Name the incentive in the invite, not the topic.
• Distractors are free red-flag detectors: anyone who selects EVERY option in a select-all list is over-eager — terminate (NN/g red-flag rule).

Exclusion-first ordering

Cheapest, most-exclusionary questions first — respondents who will never qualify exit in 30 seconds, panel costs stay down, and knockouts fire before later questions can leak the topic.

1. Conflict-of-interest knockout   — works in UX/market research, the client's
                                     industry, or a direct competitor (self or household)
2. Past-participation knockout     — paid research study in the last 3–6 months
3. Behavioral qualifiers            — masked select-all + frequency bands
4. Articulateness check             — open-ended (interviews only)
5. Quota / segmentation             — only among people already qualified
6. Logistics + consent              — device, availability, recording OK
7. Contact details                  — last, only from qualifiers

Never open with demographics or quota questions — you'd pay to classify people you're about to reject, and lead with the least masked content.

Professional-participant red flags

Professional participants are people who do studies for income. They are articulate, agreeable, experienced — and not your user. Traps to build in:

• Selects every option in any select-all list → terminate.
• Past-participation question is masked: "When did you last take part in a paid research study or focus group?" with graduated bands — not "Are you a professional participant?"
• Consistency trap: ask the same fact twice in different forms (frequency band early, "last time you…" open-ended later). Contradiction → reject.
• Open-ended answer reads like marketing copy — polished, generic, no concrete detail ("I love exploring innovative solutions") → reject. Real users name specifics.
• Implausibly broad experience — qualifies for every quota cell at once → reject.
• Incentive calibration: an incentive far above market rate for the effort actively attracts professionals. Generous, not jackpot.

Recruiting logistics

Channels — pick for representativeness, not convenience (a convenient sample answers a different question than the one in the plan). Each channel below carries a sampling bias; name and control it per name-and-control-bias — a bias that cannot touch the study's conclusion is noise, one that can needs a mitigation or a stated caveat. Convenience is a bias, not a plan:

Channel	Best for	Bias to flag
Own customer/user list	Existing-product studies, high authenticity	Misses non-users and churned users
Research panels (e.g. User Interviews, Respondent, Prolific)	Speed, niche B2B profiles	Highest professional-participant risk — traps mandatory
Live-site / in-app intercept	Catching behavior in the moment	Skews to heavy users
Community / social posts	Niche interest groups	Self-selection of the enthusiastic
Snowball (referrals from qualified participants)	Hard-to-reach populations	Network homogeneity
Friends, family, coworkers	Pilots ONLY	Never for real sessions — relationship + sample bias

Incentives — scale with session length, effort, and profile rarity. Specialist/expert profiles command multiples of general-population rates — check the current rate card of your panel rather than guessing a number. Gift cards travel better than cash across clients and schools. State amount and payment timing in the invite. Employees of the client get no cash incentive (use goodwill/swag) — cash from their employer distorts behavior.

Over-recruit buffer — no-shows are certain, not possible:

• Minimum +1 recruit per 5 sessions (20%) for moderated remote.
• In-person: recruit a floater (paid full incentive to wait on standby).
• Unmoderated remote: over-recruit 30–50% — dropout and unusable sessions are both higher.
• Confirm twice: at scheduling and 24h before. Send calendar invites with the join link.

Worked example — screener for a salon booking app usability test

Context: moderated remote usability test (60 min) of his client's salon booking app (in beta). Plan says: N=5, target = people who book personal-care appointments and manage them recurringly; quota mix of online-bookers and phone/walk-in bookers. Recruit 6 (5 + 20% buffer). Public title: "Short survey about everyday scheduling habits."

Q1. Do you, or does anyone in your household, work in any of these fields?
    (Select all that apply)
    ☐ Marketing or market research   ☐ Software or app design
    ☐ Hair, beauty, or personal-care services   ☐ Journalism
    ☐ None of the above
    LOGIC: TERMINATE if any box except "None of the above".

Q2. When did you last take part in a paid research study or focus group?
    ○ Within the past 3 months  ○ 3–12 months ago  ○ More than a year ago  ○ Never
    LOGIC: TERMINATE if "Within the past 3 months".

Q3. Which of the following have you done in the past 3 months?
    (Select all that apply)
    ☐ Ordered groceries online          ☐ Booked travel with an overnight stay
    ☐ Scheduled a personal-care appointment (haircut, salon, barber, spa)
    ☐ Sold an item on a resale marketplace
    ☐ Attended a live online class      ☐ Renewed a subscription service
    ☐ None of the above
    LOGIC: MUST include "Scheduled a personal-care appointment".
           TERMINATE if ALL options selected (red flag).

Q4. About how often do you get a haircut or other personal-care service?
    ○ Every 2–4 weeks  ○ Every 1–2 months  ○ A few times a year  ○ Rarely or never
    LOGIC: ACCEPT "Every 2–4 weeks" or "Every 1–2 months". TERMINATE otherwise.

Q5. Thinking of your most recent appointment — how did you book it?
    ○ Phone call  ○ Walked in  ○ Business's website or app
    ○ Someone else booked for me
    LOGIC: QUOTA — 3 × "website or app", 3 × "phone/walk-in".
           TERMINATE if "Someone else booked for me".
           CONSISTENCY: an online booking here with no online activity in Q3 → reject.

Q6. Briefly, walk us through the last time you booked an appointment for a
    personal service — how did you decide where and when?
    [open text]
    LOGIC: PASS = 2+ sentences with concrete detail (a place, a day, a reason).
           REJECT = blank, one word, or generic filler.

Q7. Sessions are 60-minute video calls on a phone. Are you able to join from a
    smartphone, share your screen, and be recorded for research purposes only?
    ○ Yes  ○ No        LOGIC: TERMINATE if "No".

Q8. Availability next week (select all that work): [slots]  → Contact details.

Recruiting plan: primary channel = client's customer list (invite via email); backup = one consumer panel with Q1–Q2 traps mandatory. Incentive: gift card, sized to 60 min general-population rate on the panel's current card. Recruit 6 for 5 sessions; confirm at booking and 24h out.

Anti-patterns

Don't	Do
Yes/no on the target behavior ("Do you use booking apps?")	Masked select-all with plausible distractors
Screener or invite names the product or topic	Neutral public title; topic revealed only in-session
"Would you use a feature that…?"	Past behavior only — screeners never ask predictions
Demographics as proxy ("25–40, tech-savvy")	The behavior itself, with a recency window
Distractors nobody would pick ("Wrote a novel online")	Distractors as ordinary as the target
No "None of the above"	Always include it — forced choice hides non-qualifiers
Questions with no accept/reject logic attached	Every question carries LOGIC or gets cut
Quota and demographics before knockouts	Exclusion-first ordering, always
Recruit exactly N	N + buffer (20% moderated, 30–50% unmoderated) + floater in person
Jackpot incentive to fill fast	Market-rate incentive + red-flag traps
Analyzing screener answers as research findings	Screener data gates entry only — findings come from the study
Recruiting coworkers/friends because it's Friday	Named channel with its bias flagged; convenience is a bias, not a plan

Output format

Deliver two blocks in one document:

# Screener — [neutral public title]
Study (internal): [real study name] · Profile source: [research/test plan link]
Recruit: [N + buffer] for [N] sessions · Quotas: [cells]

Q1…Qn — each as: question text, options, LOGIC line (TERMINATE / MUST / QUOTA / PASS-REJECT)

# Recruiting plan
Channel(s): [primary + backup, each with its bias flag]
Incentive: [form + basis for amount]
Buffer: [count + floater if in-person]
Schedule: [slots, confirmation cadence]
Pilot: [2–3 pass-cases + 1 fail-case, run before launch]

Sources

• NN/g — Screening Questions to Select the Right Research Participants: https://www.nngroup.com/articles/screening-questions-select-research-participants/
• NN/g — Screening Participants: https://www.nngroup.com/articles/screening-participants/
• NN/g — Recruiting and Screening Research Candidates: https://www.nngroup.com/articles/recruiting-screening-research-candidates/
• NN/g — Screen Your Research Participants to Avoid Bias (video): https://www.nngroup.com/videos/screen-your-research-participants-avoid-bias-user-research/

Boundaries

• write-research-plan owns the participant profile, inclusion/exclusion criteria, and sample size — they arrive here as input. If they don't exist, go there first.
• plan-usability-test sets profile and N for usability tests; this skill turns that into the screener and recruiting mechanics.
• write-interview-guide consumes screener answers as participant vocabulary for session questions; this skill ends when qualified participants are scheduled and confirmed.
• write-survey owns self-administered research instruments and the general per-question bias audit; a screener is a gatekeeping questionnaire — its answers are never analyzed as findings.
• name-and-control-bias owns the canonical bias→control catalog; this skill only flags per-channel sampling bias and the professional-participant red flags as local instances of it. When a flagged bias could actually reach the study's conclusion, control it per that reference — don't just note it.
• Claims about who the target users are trace to the research plan's evidence, handled per craft-critique's evidence protocol — a screener cannot repair an unevidenced profile.