UX research methods: the complete guide for 2026

Qualitative research is exploratory. It answers the "why" and "how" behind user behavior and prioritizes depth over breadth. User interviews, focus groups, diary studies, and contextual inquiry all live here. The tradeoff is smaller samples and findings that are harder to generalize, plus the risk that a researcher's own reading colors the results.

Quantitative research measures behavior through numbers. It answers "what," "where," and "how many," using metrics like task completion time, conversion rate, or drop-off. Surveys, analytics, A/B testing, and tree testing sit on this side. The strength is statistical weight you can generalize to a wider population. The weakness is that numbers rarely explain themselves.

Attitudinal research studies what people say: their stated beliefs, opinions, and intentions, gathered through interviews, surveys, and focus groups. It is useful for understanding preferences and expectations, but people are famously bad at predicting their own behavior, so stated intent and real action often diverge.

Behavioral research studies what people do: their actual actions with a product, observed through usability testing, A/B testing, analytics, and eye tracking. It surfaces problems users cannot articulate or are not even aware of. On its own, though, it can show you the "what" without the "why," which is why pairing the two works so well.

Generative research comes early, when you are still defining the problem. It generates ideas and uncovers needs through interviews, contextual inquiry, and diary studies. It is how you make sure you are solving the right problem before anyone writes code.

Evaluative research comes later, once there is something to test. It checks whether a design works through usability testing, concept testing, and A/B testing, and it hands designers specific, actionable feedback. Its scope is narrower, tied to what you have already built, so it tells you where a design fails more readily than what you should build instead.

In moderated research, a researcher is present in real time, in person or over video, and can probe, clarify, and follow a thread wherever it goes. In-depth interviews and moderated usability tests are the classic examples. You get depth and flexibility, at the cost of scheduling and time.

In unmoderated research, participants complete the study on their own, usually through an online platform. Unmoderated surveys, card sorts, and usability tests are faster, cheaper, and easier to scale to a bigger sample. The tradeoff is no chance to ask a follow-up when a participant does something surprising. Most solid research plans use both, matching depth against scale.

The fifth split, remote vs in-person, cuts across all of these. Remote research widens your participant pool and removes travel, and it is now the default for most teams. In-person research still wins when physical context and body language matter, like watching a nurse use software at a busy ward station. For more on structuring all of this, see the complete guide to research operations.

User interviews are one-on-one conversations where a researcher asks a participant about their experiences, behavior, and attitudes toward a product or problem. They are qualitative, mostly attitudinal, and generative, and they can run structured, semi-structured, or unstructured depending on how much you want the conversation to wander.

Best practices: prepare an interview guide of open-ended questions, build rapport before you dig in, stay neutral so you are not leading the witness, and record with consent so you are not scrambling to remember quotes later. Reach for interviews early, when you need to understand needs and pain points in depth. Skip them when you need statistical significance across a large group.

AI-moderated interviews are qualitative interviews conducted by an AI moderator that asks questions, listens to answers, and probes with follow-ups, all without a human researcher on the call. They are qualitative, sit between attitudinal and behavioral, and can be generative or evaluative depending on the script. Because they run asynchronously and in parallel, they let a small team gather interview-depth insight at survey-scale numbers.

This is the method that changed most in the last year. Great Question's AI Moderation (GA July 2026) runs moderated-style interviews with your own customers, then AI theme-clustering groups the responses and the generated report links every claim back to the exact clip or transcript line it came from through evidence-linked citations. Results land in the same unified results table as your surveys and usability tests, so qualitative and quantitative sit side by side instead of in separate tools.

The scale point is not hypothetical. One enterprise customer went from roughly 10,000 customer interviews a year to 100,000 by making research something the whole product org could run. Reach for AI-moderated interviews when you want the depth of a conversation but need more than a handful of them, or when scheduling live sessions is the bottleneck. Keep a human in the loop for sensitive topics and for the judgment calls AI still should not make on its own.

Focus groups are moderated discussions with 6 to 10 participants who react to a product or topic and build on each other's comments. They are qualitative, attitudinal, and generative, usually moderated, and increasingly run remotely. Group dynamics can spark ideas a one-on-one would miss, but a dominant voice can skew the room and the small sample limits how far you can generalize. Use them early for a broad read on attitudes and reactions. Avoid them for sensitive subjects or when you need individual depth.

Surveys collect structured data from a large sample through a fixed set of questions, open-ended or closed. They are usually quantitative, attitudinal, and evaluative, run remotely and unmoderated. Surveys scale cheaply and give you statistically meaningful numbers, but they rely on self-report and give you no way to ask "wait, why?" Define your objective first, keep questions clear and unloaded, and pilot the survey before you send it wide.

Diary studies ask participants to log their activities, thoughts, and experiences around a topic over days or weeks, through text, photos, or voice notes. They are qualitative, behavioral, and generative, run remotely and unmoderated. They capture behavior in real context over time, which a single session cannot, and they are the right call for understanding habits or infrequent events. The catch is participant drop-off and a pile of data to analyze, so keep the daily task light and check in regularly.

Field studies put the researcher in the participant's natural environment to observe how they actually use a product in context. They are qualitative, behavioral, and generative, and usually moderated and in person. Watching real context surfaces insights a lab never would, though coordination is expensive and the Hawthorne effect means people behave differently when observed. Use them when environment and context are central to the problem.

Contextual inquiry is a focused kind of field study where the researcher observes and interviews a participant at the same time, while they work through real tasks in their own environment. It is qualitative, blends attitudinal and behavioral, and is generative. You get the "what" from watching and the "why" from asking, in the same session. It is one of the richest generative methods available, and it is especially valuable for complex or specialized workflows where an outsider cannot guess what matters. It also takes real time to schedule, run, and synthesize, so it suits deep discovery rather than quick checks.

Card sorting asks participants to organize labeled cards into groups that make sense to them, which tells you how they mentally model your content. It can be open, closed, or hybrid, and it is qualitative, attitudinal, and evaluative, run remotely or in person. It is the go-to for designing navigation and information architecture. It will not tell you why people grouped things as they did, so pair it with a few interviews when the logic matters. Start with a card sorting template.

Tree testing, sometimes called reverse card sorting, gives participants a task and a text-only version of your site structure and checks whether they can find what they need. It is quantitative, behavioral, and evaluative, run remotely and unmoderated. It isolates structure from visual design, so you learn whether your information architecture works before a single pixel is designed. Pair it with card sorting: card sorting helps you build the structure, tree testing validates it.

Usability testing watches representative users attempt real tasks with your product while you observe where they struggle. It is both qualitative and quantitative, behavioral, and evaluative, run moderated or unmoderated, in person or remote. It is the most direct way to find friction in an actual interface. It tells you where users get stuck, not what they wish existed, so keep it evaluative. Set clear goals, recruit users who match your real audience, and iterate after you fix what you find.

Prototype testing puts an early model of your product, from a rough sketch to a clickable mockup, in front of users to catch problems before you build. It spans qualitative and quantitative, is behavioral and evaluative, and runs moderated or unmoderated. It saves you from expensive redesigns after launch, though low-fidelity prototypes sometimes need a facilitator to fill in the gaps. Match the fidelity to the question, and help rather than lead. Grab a prototype testing template to start.

Five-second testing shows participants a design for five seconds, then asks what they remember and what they thought the page was for. It is quantitative and qualitative, attitudinal, and evaluative, run remotely and unmoderated. It measures first impressions and whether your core message lands instantly, which matters most for landing pages and hero sections. It only captures the first few seconds, so it complements deeper usability work rather than replacing it.

First-click testing checks where users click first when given a task, on the logic that a right first click strongly predicts task success. It is quantitative, behavioral, and evaluative, run remotely and unmoderated. It is a fast, cheap way to validate whether your layout points people toward the right action. It tells you about that first decision only, so use it alongside full-task usability testing when the whole flow matters.

Concept testing puts an early idea, feature, or value proposition in front of your target audience to gauge appeal and understanding before you invest in building it. It can be qualitative or quantitative, is attitudinal, and spans generative and evaluative. It helps you kill weak concepts early and sharpen strong ones, and it works well as a survey, an interview add-on, or a preference test between options. Because it measures stated interest, confirm the signal with behavioral testing once something real exists.

A/B testing shows two versions of a design to different users at random and measures which performs better on a real metric like conversion or click-through. It is quantitative, behavioral, and evaluative, run remotely and unmoderated. It removes guesswork from decisions between two options, but it only compares what you already have and needs enough traffic to reach significance. Change one variable at a time, and use it to optimize existing designs rather than to discover new ones.

The bottleneck for most teams is not knowing the methods. It is running them without a tangle of separate tools for recruiting, scheduling, moderating, and analysis. ServiceNow cut participant recruitment from 118 days to 6 and consolidated 15 tools down to 7 by moving research onto one platform. Asana compressed research cycles from two weeks to a few days. Running your methods against your own customers, in one place, is what makes a mixed-methods plan actually feasible instead of aspirational.

Great Question brings recruiting, moderated and AI-moderated interviews, surveys, card sorts, prototype tests, usability tests, and a shared research repository and synthesis into a single platform, run against your own customers rather than a panel of paid testers. That is the difference between choosing a method because it fits the question and choosing one because it is the only thing your tooling makes easy.

Ready to run these methods against your own customers instead of a Frankenstein of tools?