How many user interviews are enough?

There's no single number for how many user interviews you need. It depends on how similar your participants are, how focused your question is, and whether you're comparing segments. As a rule of thumb, about 12 interviews with a fairly homogeneous group on a focused question usually surface most of the main themes; when participants vary widely or you're comparing segments, plan each segment separately. After reading this, you'll be able to set a sample size you can justify for your next study, and know when to add participants and when to stop during the interviews.

Thematic saturation: how to tell you've done enough interviews

In qualitative research, the usual standard for whether a sample is large enough is thematic saturation: interviewing new participants rarely produces new themes, and what you hear mostly repeats or adds to existing ones. At that point, the return on adding more people drops sharply.

The most cited study is Guest, Bunce & Johnson (2006). They analyzed 60 in-depth interviews, recording newly emerging codes interview by interview, and found that in a relatively homogeneous group, the first 12 interviews covered the vast majority of themes, and the main high-level themes had already appeared within the first 6.

When citing this study, keep its conditions in mind: participants had similar backgrounds, the interview guide was fairly structured, and the questions were narrowly focused. If your participants span new and existing users, several countries, or multiple pricing tiers, or your research question is very open, 12 interviews won't be enough. It's a starting point, not a rule.

Saturation applies to a specific group and a specific question. If the group or the question changes, you need to judge it again.

5 users for usability testing: where it applies and how it's misused

In a 2000 article, Jakob Nielsen argued that testing with 5 users in a round of usability testing uncovers most of an interface's usability problems, and that rather than testing 15 people at once, it's better to run three rounds of 5, fixing issues between rounds. This rests on one premise: everyone completes the same tasks on the same interface, so the obstacles they hit overlap heavily.

A common misuse is applying "5 is enough" to all user research. Exploring needs, understanding motivations, and comparing attitudes across groups produce far more varied answers, and 5 interviews often surface only the most obvious views. In the same article, Nielsen also notes that when user groups differ significantly you should test them separately, and that quantitative studies that need statistics require larger samples.

How many participants per group when comparing segments

If one goal of the study is to compare groups, such as churned vs. retained users, new vs. experienced users, or different cities, saturation has to be reached within each group. With 12 people split into three groups of 4, no difference between groups will hold up.

• First list the groups you must compare, usually no more than 3. The more groups you have, the thinner each one gets.
• Size each group as you would a homogeneous group: usually start at 6 to 8, and move toward 12 when the question is open or the group varies a lot internally.
• For dimensions where you just want broader coverage and don't plan to compare groups, use quotas to make sure each type has at least a few people. You don't need a full group for each.
• When one group is especially hard to recruit, narrow the comparison rather than let two or three people stand in for a whole group.

This table is a rule of thumb, not a statistical requirement. The real number depends on whether new themes keep appearing in your interviews.

During interviews: when to add participants and when to stop

Set a planned sample size before you start, and adjust it during the study based on what you see. Pause after every 3 to 4 interviews, log new themes in a table, and see whether the curve is leveling off.

Signs you should add participants

• Recent interviews keep surfacing new themes, with no sign of leveling off.
• A key finding is supported by only one or two participants and will directly affect the decision.
• Unexpected differences between groups show up, such as users from one channel saying noticeably different things.
• Participants in one group clearly skew one way, for example all active users, with no quiet or churned users.

Signs you can stop

• About 3 interviews in a row bring no new themes; new material only adds detail.
• Each research question has quotes from several participants backing it, and you've found counterexamples.
• More interviews won't change the decision you need to make.

AI interviews allow larger samples: what changes and what doesn't

AI-moderated interviews can run many sessions at once at a lower cost per session, making it practical to grow your sample from a dozen or so participants to dozens or even hundreds. This changes a few things in practice: you can fully staff every group, pull out edge groups to hear from separately, and back your findings with more meaningful counts.

Some things haven't changed. Sample representativeness still depends on recruiting: 100 interviews with only active users are worth less than 20 that cover new, existing, and churned users. Interview quality still depends on the guide and follow-up questions; if the questions stay shallow, more people just repeat the same layer of information. With more participants, synthesis takes more work, which makes it even more important to group by theme, note counts, and keep quotes. Counts from qualitative interviews also shouldn't be treated as statistical findings; when you need precise proportions, validate them with a survey or behavior data.