UX Research Methods

Choosing the method = knowing your question. Two axes (Rohrer/NN/g landscape): attitudinal↔behavioral, qualitative↔quantitative. Cardinal rule: what people SAY ≠ what they DO — behavioral evidence trumps stated preference for behavior questions.

Discovery (what to build)

Semi-structured interviews: ask about past behavior, not hypothetical futures ("walk me through the last time you…" not "would you use…"). Five whys for causes. Sample-size guidance: Guest et al. (2006) found most major themes appeared within ~6 interviews and code saturation by ~12 in a homogeneous sample; the common "5–8 per segment per round" heuristic is a practitioner convention derived from that kind of finding, not a law — heterogeneous populations need more.
Contextual inquiry / field observation (Beyer & Holtzblatt): watch work happen where it happens — surfaces workarounds and unspoken needs interviews miss.
Diary studies: longitudinal, in-context experience sampling for episodic behaviors.
Surveys: quantify KNOWN issues (bad at discovering unknowns); pilot-test wording; avoid leading/double-barreled questions (questionnaire-design canon: Dillman; Tourangeau).
Analytics/support mining: where do users actually struggle, drop, rage-click, contact support — free behavioral discovery.

Continuous discovery (Torres)

Torres's definition: at minimum weekly touchpoints with customers, by the team building the product, doing small research activities in pursuit of a desired product outcome — discovery as a habit, not a phase gate before build.
Supporting practices: automate recruiting (in-product intercepts, standing panels); interview snapshots so learning compounds; the opportunity solution tree maps outcome → opportunities (from interviews) → solutions → assumption tests, keeping solutions tied to evidenced needs rather than pet ideas.

Structure (how to organize it)

Card sorting (open/closed): users group content → informs IA categories. Tree testing: validates a hierarchy by findability tasks on the bare structure. (Both: Spencer; see Information Architecture.)

Evaluation (does it work) — see Usability Testing & UX Metrics for detail

Moderated/unmoderated usability tests, heuristic evaluation, A/B tests, first-click testing. On the famous first-click statistic: Bailey & Wolfson's 2006–2009 analyses of live-site tasks reported ~87% task success after a correct first click vs. ~46% after a wrong one — but this is practitioner gray literature (webusability.com), never peer-reviewed, and MeasuringU's later replication found a much smaller (though still real) advantage. Treat "first click matters" as directionally solid and the 87/46 figures as one dataset, not a constant.

AI-assisted and AI-moderated research (state of practice, mid-2026)

AI as analyst's assistant (mature enough to use with care): transcription, clustering candidate themes, drafting summaries. Validity rule: the researcher verifies every claimed theme against raw data — LLMs paraphrase plausibly and can flatten or invent nuance.
AI moderators (emerging): platforms run scripted interviews with dynamic follow-ups. NN/g's hands-on evaluations find them serviceable for structured, consistency-first studies, screening, and multilingual reach — and weak at probing emotional subtext or pivoting when a participant reveals something unexpected. Scale without rigor just produces flawed research faster.
Synthetic users (skepticism warranted): LLM-generated "participants." NN/g's comparison against three real studies (Rosala & Moran, 2024) found responses shallow, one-dimensional, and sycophantic — praising concepts real users rejected. Usable at most for brainstorming hypotheses to test with humans; not a substitute for research with real users. Academic work on LLM personas is active but unsettled.

Synthesis, repositories, and traceability

Affinity diagramming from raw observations up to themes (bottom-up); personas/JTBD/journey maps as OUTPUTS of data, never inputs invented in a workshop; every insight tagged to its evidence (traceability keeps stakeholder debates factual).
Research repositories (Dovetail-style tools or a disciplined wiki): store atomic, evidence-linked insights — the "atomic research" framing (Pidcock): experiment → fact → insight → recommendation. Conventions that make repositories survive: consistent tagging taxonomy, an owner, and expiry/review dates — insights about a product decay as it changes.
Mixed-methods triangulation: converge at least two independent method families before big bets — qual explains the "why" behind quant patterns; quant sizes the qual finding; agreement across attitudinal + behavioral sources is the strongest signal you have.

Informed consent covers purpose, recording, data use, and retention — obtained before the session, re-confirmed on camera for recordings; participants may withdraw (and have their data deleted) at any time.
PII minimization: collect only what the study needs; pseudonymize in the repository; store recordings access-controlled with a deletion schedule (GDPR/CCPA apply to research data like any other).
Incentives: pay fairly for time (market-rate honoraria); avoid amounts so large they become coercive for vulnerable populations; pay even if the session fails for technical reasons.
No deception without debrief; never test real credentials/finances; screen-out respondents get a courteous exit, not a trick ending.

Rigor guardrails

Recruit real target users (not colleagues); separate moderator/notetaker; neutral task wording (no interface vocabulary in the task); pilot every study; distinguish observation ("7 of 8 missed the link") from interpretation ("the label lacks scent"); triangulate methods before big bets.

Sources

Rohrer, C. — "When to Use Which UX Research Methods," NN/g (nngroup.com).
Beyer, H. & Holtzblatt, K. (1998). Contextual Design. Morgan Kaufmann.
Portigal, S. (2013). Interviewing Users. Rosenfeld Media.
Guest, G., Bunce, A. & Johnson, L. (2006). "How many interviews are enough?" Field Methods, 18(1).
Torres, T. (2021). Continuous Discovery Habits. Product Talk LLC.
Spencer, D. (2009). Card Sorting. Rosenfeld Media.
Goodman, E., Kuniavsky, M. & Moed, A. (2012). Observing the User Experience (2nd ed.). Morgan Kaufmann.
Bailey, B. & Wolfson, C. — first-click analyses (webusability.com, gray literature); smaller-effect replication at measuringu.com.
Rosala, M. & Moran, K. (2024). "Synthetic Users: If, When, and How to Use AI-Generated 'Research'." NN/g (nngroup.com/articles/synthetic-users/).
NN/g — "Accelerating Research with AI" and AI-moderation evaluations (nngroup.com).
Pidcock, D. — "Atomic UX Research" (practitioner framework).