How many users do you need for user research?
Five to eight users per segment for qualitative usability work — that range surfaces most severe issues, and returns drop sharply after. Quantitative claims need hundreds. The mistake isn't sample size, it's using five interviews to justify a number instead of a direction.
The two questions are different
Qualitative asks why and how — where people get stuck, what they misunderstand, what they were actually trying to do. Small samples work because obstacles are shared: if three of five people can't find the export button, the sixth won't change your mind.
Quantitative asks how many and how much. "Does variant B convert better?" needs enough traffic for the difference to be distinguishable from noise, which is usually thousands of sessions, not five.
Most research arguments are really an unstated disagreement about which question is being asked.
Segment first, then count
Five per segment, not five total. If your product serves both a solo practitioner and an enterprise admin, five sessions split across both tells you little about either — you have two-and-a-half users per group, which is anecdote.
Intercom's discovery practice reflects this: the job people were hiring the product for varied sharply between segments, so blending them would have averaged away the very distinction that mattered.
Recruit for the situation, not the demographic
The most common recruiting error is screening on job title. What determines behaviour is whether someone is currently in the situation your product addresses — someone who tried to solve this problem in the last month will tell you more than someone with the right title who has never had the problem.
Headspace's early design work leaned heavily on a specific struggle: people who wanted to meditate and found existing tools intimidating. That's a situation, not a demographic, and it's a far better recruiting screen.
Small samples, honest claims
Five interviews justify "we saw people struggle here, let's fix it." They don't justify "62% of users prefer this." Use the evidence at the strength you gathered it, and the sample-size argument mostly disappears.
Seen in practice
Case studies where this shows up as a real decision, not a definition.
Related questions
Where does the five-user rule come from?
From Jakob Nielsen and Tom Landauer's work in the early 1990s, modelling problem discovery against participant count. Five users typically surface around 80% of usability issues in a single interface for a single user type — a figure that assumes one segment and one task flow.
When is five users not enough?
When you have distinct segments, when the flows diverge substantially, or when the question is about preference or magnitude rather than obstacles. Five users cannot tell you which of two designs converts better; that requires a test with statistical power.
Should you keep interviewing after you stop hearing new things?
No — saturation is the practical stopping rule. When two or three consecutive sessions produce nothing you haven't already heard, additional sessions are confirming rather than discovering, and the time is better spent acting on what you found.
More on discovery
Last reviewed 2026-09-07