UX Research
Methods:
The Complete Guide

One-on-one conversations with users or potential users. The goal is to understand motivations, mental models, and the context behind behavior — not to test anything specific. Best run with a topic guide rather than a fixed script: let the participant lead into unexpected territory. A sample of 5–8 participants typically reaches thematic saturation for a bounded question. Output: rich, verbatim evidence that grounds every subsequent decision in real human context.

Observation in the user's real environment — workplace, home, commute — while they perform actual tasks. The researcher observes and asks contextual questions ('what just happened?', 'why did you do that?') without directing the work. Contextual inquiry surfaces workflow-level problems that no interview or usability test can find: the workarounds, the informal tools, the context switches. It is expensive but produces insight that no other method delivers.

A participant attempts realistic tasks using a prototype or live product while a researcher observes and probes in real time. The moderator's power is the follow-up question: when a user hesitates, fails, or does something unexpected, the researcher can immediately ask why. Five participants per target user segment typically surfaces 80%+ of critical usability problems. Output: specific, reproducible failure modes with direct evidence of where and why they occur.

Participants complete tasks independently, typically via a platform tool, while screen and think-aloud are recorded. No researcher is present. This allows 20–100+ participants to run overnight at a fraction of the cost of moderated sessions. Unmoderated tests excel at validating whether a specific flow works across a wider and more representative sample, and at detecting severity — whether a problem is isolated or widespread. They cannot explain unexpected behaviour without a follow-up.

Concept tests evaluate ideas, directions, or early-stage designs before significant investment. They can be moderated (show three directions, discuss) or unmoderated (preference questions, first-click tasks, desirability scales). The goal is directional signal: which concept has the strongest resonance, the clearest value proposition, the fewest barriers. Best combined with qualitative probing to understand the 'why' behind stated preferences.

Structured questionnaires delivered to large samples — typically 50–500+ respondents depending on the statistical confidence required. Surveys answer prevalence questions: how widespread is this problem, this belief, this behaviour across the user base? They are the only method that generalises findings across a population. Their limitation is context: surveys cannot explain why respondents said what they said. Best used after qualitative research has identified the right questions to measure.

Participants sort topic cards into groups and name those groups (open sort) or sort into predefined categories (closed sort). Card sorting reveals the user's mental model of how information should be organised — critical input for IA, navigation labels, and content hierarchy. Open sorts surface emergent categories; closed sorts validate whether a proposed taxonomy matches user expectations. Output: dendrograms and similarity matrices that quantify grouping patterns across participants.

Participants find items in a text-only version of the navigation hierarchy (no visual design). Tree testing isolates whether the information architecture itself is findable, removing all visual and labelling noise. Directness score (found without backtracking) and success rate per task are the core metrics. Run after card sorting to validate whether the proposed IA structure works before committing to design. Results are directly actionable: branches with low success rates need restructuring.

A/B tests expose two versions to separate user segments and measure performance (conversion, completion, engagement) with statistical significance. Preference tests ask users to choose between designs and explain why. A/B tests tell you which version performs better; preference tests tell you which version users say they prefer. The two should agree — when they don't, investigate the discrepancy rather than trusting either in isolation. Both methods require clear hypotheses and adequate sample sizes before running.

Participants log experiences, tasks, or product interactions over days or weeks, often prompted by the research team at set intervals. Diary studies capture longitudinal context: how behaviour evolves over time, how the product fits (or fails to fit) into real-world routines, and how first impressions shift with repeated use. They are uniquely suited to discovering habit formation, workarounds, and abandonment patterns that no single-session method can see.

Structured analysis of quantitative behavioural data — pageviews, funnels, session recordings, click maps, retention curves. Analytics tells you exactly what users do at scale but cannot tell you why. The practical workflow: identify anomalies in analytics (drop-offs, rage clicks, unexpected paths), then design qualitative research to explain them. Analytics is also the fastest way to detect the downstream impact of a design change — it is evidence of behaviour, not opinion about behaviour.

Participants view a design for exactly five seconds, then answer questions about what they remember and what the product does. Five-second tests measure first impressions and value-proposition clarity: can users understand what this product does, for whom, within a single glance? They are fast to run and cheap, making them useful early in the design process to validate that a landing page, onboarding screen, or key feature communicates its purpose clearly before investing in more elaborate evaluation.