Running Focus Groups for UX Research: What They Answer and What They Wreck

Eight people sit around a table. The moderator shows two versions of an onboarding flow and asks which one is clearer. One participant — confident, early, articulate — picks the second. Two people nod. By the time the question comes round the table, six of eight prefer version two, and the report says so.

Nobody in that room used either version. They looked at pictures of them and agreed with a stranger.

That study cost real money and produced a number the team will quote for a year. The method wasn't broken. It was pointed at a question it structurally cannot answer.

What a focus group actually is

A focus group is a moderated discussion among six to eight participants, usually ninety minutes, built around a topic rather than a task. The unit of data is the conversation: what people say to each other, which words they reach for, where they disagree, what they assume everyone already knows.

That last part is where teams go wrong. Focus groups are a group method. Whatever comes out belongs to the room. Run eight separate interviews instead and you get eight independent accounts, each one deeper than anything the group produced.

Teams book them as cheaper interviews run in parallel. They are a separate instrument with one reading on the dial: how a topic gets talked about when people talk about it together.

The narrow band where they're genuinely good

There is a real job here, and it sits upstream of design.

Language and vocabulary. What do people call the thing you built? A room of eight will surface five competing terms in ten minutes and then argue about which is right, and that argument is unavailable from a survey and would take weeks to assemble from interviews. If your nav labels, empty states, and pricing page are guessing at a user's word for something, this is the fastest way to stop guessing.

Framing and positioning. Show a value proposition and let people react to each other's reactions. Objections surface in a group that politeness suppresses one-on-one, because the first skeptic gives everyone else permission.

Assumed context. Focus groups are unusually good at exposing what a community treats as too obvious to mention. When four people in a row skip past the same step in describing their workflow, you've found something they all consider background — often a workaround so ingrained they no longer see it.

Range-finding, before you know the questions. Early, when you can't yet write a decent interview guide, an hour with a group tells you what the guide should ask about.

Every item on that list is about talk. Words, frames, shared assumptions, unknown unknowns. Whether someone can complete a task is a different question, answered by a different method.

What they wreck

The failure mode is consistent enough to name: teams use focus groups to evaluate designs, and the group produces an opinion that fails to predict behavior.

Jakob Nielsen has been blunt about this for two decades — group discussion is a poor way to evaluate usability, because what people say about an interface in a room has little relationship to what they do when they're alone with it and actually trying to get something done. The gap between stated preference and observed behavior isn't a sampling problem you can fix with more groups. It's built into asking.

What makes it worse is that the distortions compound. Someone answers first, confidently, and that answer becomes the frame everyone else either agrees with or corrects — so the room's output is partly a function of seating and temperament. Then peer presence kicks in: people perform for each other, over-reporting the effortful, principled choice (the budgeting app they'd totally use, the privacy setting they'd definitely configure) and quietly dropping the lazy real one. That performance pressure pushes toward agreement, because disagreeing with seven strangers is socially expensive, so genuine polarization gets sanded into a moderate group position describing nobody present. And running underneath all of it, participants are reading the moderator for what a good answer looks like — one person guesses right, gets a nod, and the rest calibrate to the nod.

Each of those can be dampened. None can be removed, because they are what a group is.

Match the method to the question

Before booking anything, say the question out loud and check which instrument it belongs to.

Question	Method
Can someone complete checkout without help?	Usability test
Why did they abandon it?	1:1 interviews, session review
What do people call this feature?	Focus group
Which of two headlines performs better?	A/B test
How widely does this complaint hold?	Survey
What objections will sales hit?	Focus group
Does version A beat version B on ease?	Usability test with a defined dependent variable

The pattern: focus groups answer questions about talk; behavioral questions need behavior. The other methods and the two axes that sort them are laid out on the UX research methods hub, worth five minutes before you commit budget to any of this.

Running one so the transcript is worth reading

If the question passed the test above, the execution details matter more than they look.

Screen on behavior. Age-and-income quotas buy you eight parallel monologues. Eight people who use the same category of product weekly have a real conversation, because they share enough context to disagree about something specific. Screen on recency, frequency, and role, and be ruthless about professional respondents who have done this six times this year.

Six to eight people, ninety minutes. The arithmetic sets the ceiling: subtract intro and wrap-up from ninety minutes and eight participants get roughly eight minutes of airtime each, which the two most talkative will halve for everybody else. Below six, the discussion stalls whenever a couple of people go quiet. And a single group gives you n=1 group, not n=8 participants — the room is the unit. Budget for three or four groups per segment before you treat any pattern as real.

Write first, then talk. Before opening a question to discussion, have everyone write their answer privately for a minute. Then go round. This is the most useful thing a moderator can do, and it isn't a matter of taste. Diehl and Stroebe ran the experiment in 1987 (Journal of Personality and Social Psychology) and found the same people, working alone and pooling afterward, out-produced themselves as a group. Production blocking and evaluation apprehension did the damage. Nominal-group technique, from Delbecq and Van de Ven's 1971 work on group decision-making, exists to buy that loss back. Writing first captures independent positions before the room has one.

Give them something to do. Card sorts, ranking exercises, "circle the word you'd expect to see here." A ranked stack of nine cards survives the trip to a spreadsheet. "I guess I'd probably use it?" dies in the transcript.

Let the disagreement run. When two people split, the instinct is to resolve it and move on. Resist. The reason for the split is usually the finding. Whatever consensus the room reaches afterward is mostly noise.

Ask about the past. "Tell me about the last time you needed this" produces recallable specifics. "Would you use this?" produces a prediction, and people are bad at predicting their own behavior. Whatever you get is a hypothesis that still has to be tested against something real, which is the inductive-to-deductive arc of any research program.

Online focus groups

Remote changes the physics enough to plan around.

Turn-taking collapses first. Video latency makes natural interruption impossible, so the discussion serializes into round-robin and you lose the cross-talk that made the method worth running. Working around it means putting people in breakout pairs, running a shared whiteboard so several can act at once, and cutting group size to five or six now that airtime is scarcer. You also lose the body language that tells a moderator someone disagrees but hasn't said so, and any chance of handing people physical stimulus.

Against that: no travel, wider geography, and a chat channel that quiet participants will use when they won't unmute. Treat chat as a second data stream. Dissent often shows up there first, from exactly the people the room is steamrolling. Recruit for the format and don't pretend the transcript is equivalent to being there.

Analyzing without over-claiming

The analysis trap is counting. A transcript where six of eight said something looks like 75% and reads like consensus, when it's one room, one first speaker, one afternoon. Report it as "the dominant framing across three groups was X, with a persistent minority position Y," and note who spoke first if the pattern followed them.

Code for the things the method is good at: recurring vocabulary, points where the room split, moments where someone's example changed other people's answers, and anything treated as too obvious to explain. Those are the categories where a focus group earns back its cost.

Then convert findings into testable claims. A focus group's proper output is not a decision — it's a better-informed hypothesis and a sharper vocabulary for the next study.

FAQ

What is a focus group in UX research?

A moderated group discussion, typically six to eight participants for about ninety minutes, used to explore attitudes, vocabulary, and reactions to a concept. It's exploratory and qualitative. The data is the conversation, which makes the room your unit of analysis.

What's the difference between a focus group and a user interview?

An interview isolates one person's experience and can follow their reasoning deep. A focus group deliberately introduces peer influence, which surfaces shared language and objections while contaminating individual opinion. Reach for interviews when you need someone's actual reasoning. Reach for a focus group when you need to hear how the topic gets discussed.

Are focus groups good for usability testing?

No. Discussing an interface and using one are different activities, and stated preference in a room predicts task performance poorly. Run a usability test with individuals attempting real tasks. A focus group can tell you what to call a button. Finding out whether anyone spots it takes a test.

How many focus groups do I need?

Plan for three or four per audience segment. One group is a single data point, entangled with who happened to be in it and who spoke first. Patterns that survive across three independently recruited groups are the ones worth acting on.

How is a market research focus group different from a UX one?

Mostly in the question. Market research focus groups probe purchase intent, positioning, and brand perception; UX focus groups probe workflows, terminology, and where a product fits into what someone already does. Same moderation craft, same distortions.

Are online focus groups as good as in-person?

For vocabulary and reactions, close enough. For cross-talk and body language, no — video latency serializes the conversation and removes the disagreement you'd have read off someone's face. Compensate with smaller groups, breakout pairs, shared whiteboards, and by treating the chat channel as real data.