Last updated: 1 September 2026
A focus group is the right instrument when the thing you need to observe is the interaction itself: people building on each other's language, disagreeing, negotiating what is normal in a category. It is the wrong instrument when you need an individual account that nobody else in the room has shaped. Most arguments about focus groups are really arguments about which of those two things the study needed, decided after the fieldwork rather than before it.
The method has been declared obsolete for two decades and has not obliged, which is worth taking seriously rather than dismissing. One thing has genuinely changed. Groups used to be the default qualitative format because they were the efficient way to see many people quickly, and they are no longer that. When the group is chosen now, it should be chosen for the interaction, because everything else it used to offer is available elsewhere at lower cost.
Where Does the Group Format Genuinely Outperform?
Four things a group does that an interview cannot.
- It shows you language forming. Participants reach for words to describe a category and refine them against each other. That negotiated vocabulary is frequently the most commercially useful output of a group, because it is the language a market will actually use.
- It surfaces disagreement as disagreement. In separate interviews you get two contradictory accounts and have to infer the tension. In a group you watch the tension play out, including which position people back down from.
- It reveals what is socially sayable. What a group will and will not endorse out loud is real information about a category norm, particularly for anything status-adjacent.
- It is efficient for reaction to stimulus. Showing a concept, pack or ad to eight people at once and watching first reactions is fast, and first reactions are the part least damaged by the group setting.
What Group Dynamics Do to Your Data
The same mechanism that produces the value produces the distortion, and it is not optional.
A group produces a collective account, not eight individual ones. Dominant participants move the room. Early speakers anchor later ones.
Minority views are underreported, not because people hide them deliberately but because the cost of holding an unpopular position in front of strangers is real. Anyone treating group transcripts as eight independent data points has already made an error the analysis cannot recover from.
The evidence on how this affects disclosure of sensitive material is genuinely mixed rather than settled, and it is worth knowing that before designing around an assumption. Some comparative work finds individual interviews substantially more likely to elicit socially sensitive themes; other randomized comparisons find certain sensitive disclosures occurring more readily in groups, where hearing another participant raise something appears to license it. The honest summary is that the direction depends on the topic and on who is in the room, which argues for piloting both when a study hinges on sensitive material rather than assuming the interview is always safer.
Professional expectations for participant treatment, consent and disclosure in either format are set out in the ESOMAR code and guidelines and the AAPOR standards and ethics materials.
How Many Groups or Interviews Do You Actually Need?
Fewer than most buyers assume, and the published work here is unusually specific.
An analysis in PLOS One proposing a method to assess and report thematic saturation applied bootstrapping to three interview datasets of 40, 48 and 60 interviews. Using a threshold of no more than 5 percent new information across a run of two, saturation was reached at six interviews in two of the datasets and at eight to nine in the third. The same paper notes earlier work in which 70 percent of 114 identified themes appeared in the first six interviews and 92 percent within the first twelve.
Two cautions before that becomes a budget cut. Those figures describe relatively homogeneous groups with narrow research aims; heterogeneity and broad objectives both push the number up. The federal review of participant numbers in usability research names group count, group heterogeneity and complexity as the same three inflators. And saturation within one segment or market says nothing about themes that only appear when segments are compared, which is a separate and larger sample question.
The practical translation for groups: a small number of well-composed groups per distinct segment, rather than a large number of groups covering a blurred population. Composition does more work than count, which is the same discipline that separates a usable segmentation from a shelved one in customer segmentation research.
Focus Group, Interview or Something Else?
| Format | Best at | Worst at | Typical scale |
|---|---|---|---|
| Focus group | Language formation, norm negotiation, stimulus reaction, observed disagreement | Individual accounts, sensitive or deviant behavior, quantifying anything | 6 to 10 people, several groups per segment |
| In-depth interview | Uncontaminated individual narrative, sequence of a decision, sensitive material | Seeing interaction; cost per person is high | Saturation logic, often single digits per homogeneous segment |
| Dyad or triad | Household or joint decisions, some group effect with less dominance | Broad category exploration | 2 to 3 people |
| Asynchronous online community | Behavior over time, diary-style evidence, reflection | Immediacy, spontaneous reaction | Days to weeks |
| Survey with open ends | Scale and quantification | Depth, follow-up, why | Hundreds and up |
Best for a decision about a joint purchase: the dyad, and it is chronically underused. A category bought by a household is misrepresented by both formats above it, since the individual interview misses the negotiation and the group of strangers reproduces a negotiation that never happens with strangers.
What Changed When Groups Moved Online
Video groups removed the facility cost, the travel and the geographic constraint, and they are now the default for most commercial work. The same economics now push teams toward AI-moderated interviews against human ones. Three things did not survive the move intact.
Physical stimulus. Handling a pack, feeling a fabric weight, judging a product's real scale on a shelf. Nothing on video replaces this, and studies where the tactile response is the point still justify a room.
Peripheral observation. In a facility a moderator reads the participants who are not speaking. On a grid of video tiles that signal is heavily degraded, which shifts the moderator's job toward explicit verbal checking.
Even participation. Online turn-taking is coarser than in-person turn-taking, so dominance effects tend to increase rather than decrease, contrary to a common assumption that the format equalizes.
What improved is real too: recruitment reach beyond a facility's catchment, lower cost per group, and recordings that are searchable rather than sitting on a tape.
Who Never Makes It Into the Room?
Every qualitative format inherits the reach of its recruitment channel, and the video group inherits a narrower one than the facility it replaced. Where those respondents are actually sourced is covered in survey panels and where respondents come from.
A video group requires a device with a working camera, a connection stable enough to hold it for ninety minutes, somewhere private enough to speak freely, and enough comfort with the format to talk in front of strangers on screen. Pew Research Center's mobile technology fact sheet reports 16 percent of US adults as smartphone-only internet users. Pew puts that at 34 percent among adults in households under $30,000 a year against 4 percent above $100,000, and the ITU's connectivity statistics show the gradient running considerably steeper across markets.
That is a sampling problem disguised as a logistics preference, and it points the same direction as every other selection effect in qualitative recruitment: toward the comfortable, the connected and the confident.
Where a study has to reach past that, the channel is the variable. Alchemic fields interviews natively inside WhatsApp with no link and no app to install, accepting text or voice notes and handling code-mixed speech, and by outbound phone call for respondents reachable by voice rather than browser. Alchemic publishes 57+ languages including Hindi, Tamil, Telugu, Bangla, Arabic and Indonesian.
Recruitment runs as managed fieldwork or bring your own, across fourteen markets that include the USA and the UK, and in India from metros and Tier 1 through Tier 2 and Tier 3.
None of that reproduces a group, and it is not meant to. It changes who is available to be interviewed individually when a group would have excluded them, which is the reach question examined in sample validity and who you miss.
For studies where the interaction genuinely is the point, AI-moderated interviews are a complement rather than a substitute: they scale the individual accounts, and the group still does the thing only a group does.
Where Focus Groups Are the Wrong Tool
Five limits, stated plainly.
Anything requiring an uncontaminated individual account. Purchase decision sequences, personal finances, health, anything where one participant's framing will reshape everyone else's recall.
Anything you intend to count. Six of eight people agreeing in a group is not 75 percent of a market, and it is not even six independent agreements. Reporting group counts as proportions is the most common way qualitative findings are misused downstream.
Low-incidence or highly specialist populations. Assembling eight rare respondents at one time costs more than interviewing them separately and gains little.
Categories with strong social desirability pressure. Anything where the socially approved answer is well known will produce that answer, competently and unanimously.
Testing a proposition you have already decided to launch. A group will find something to say about anything. If no result would change the decision, the money is buying reassurance.
A sixth case deserves separating out, because it is the one most likely to be defended in a meeting. Using groups to reach a population the recruitment channel cannot properly cover, on the grounds that a group is cheaper per person. Composition is what makes a group work, and a group assembled from whoever was reachable is not a segment, it is a convenience sample with a moderator.
Weighting a recruitment profile against a public frame such as US Census population data will usually reveal that skew before fieldwork rather than after. So will checking the sample's financial profile against the Federal Reserve's survey of household economics and decisionmaking. Neither check costs anything, and both are more informative than adding another group.
Teams weighing which qualitative format the decision actually needs will get more from the qualitative research primer and from the concept testing overview than from another round of format advocacy.