It is not a rule. It is a shortcut I use to work out how many people to talk to — and it only works if you do the part nobody quotes.
I already know how this number will be used against me. Someone will say "Nina says five interviews is enough", a product manager will run five conversations with whoever answered first, and a roadmap will be built on it. That person did not read the book. Let me put the part they skipped where it can't be missed.
Qualitative research has no sample-size formula. There is no equivalent of the confidence interval that tells you when to stop. What there is instead is saturation: the point at which new conversations stop producing new categories — new reasons, new patterns, new ways the thing goes wrong. Three, five and seven are the numbers at which I have learned to expect that point to arrive for a single, reasonably homogeneous group of people talking about a single, reasonably bounded question. Three tells you whether you're asking the right thing. Five usually shows you the shape. Seven tends to confirm it or break it.
That's all they are. Expectations, from experience, about where saturation tends to land. Not a rule.
The numbers only mean something after you've done three things. Skip any one and the number is just someone else's laziness with my name on it.
Decide what has to differ. The three-five-seven expectation is per group, and groups are defined by what you believe might make the answers different. If you are studying a subscription product and you have reason to think new users and long-term users cancel for different reasons, that is two groups. If you also think it matters whether they came from ads or from a friend, that's four. Three age bands and two genders is six. Now multiply: six groups, three to five conversations each, and your "five interviews" has become eighteen to thirty. The number was never five. Five was the unit.
Decide what saturation looks like before you start. Saturation is not a feeling of having heard enough. It is an observation: the last two conversations added nothing you hadn't already coded. That requires coding as you go, not after — a running list of the categories you've heard, updated after every session, so you can see the list stop growing. I keep a saturation tracker for every study. It's a boring spreadsheet. It is also the only honest answer to "how do you know you've done enough".
Look for the person who breaks it. Before you stop, deliberately go find the case that would ruin your picture — the long-term user who churned for a new-user reason, the enthusiast who cancelled anyway. If you can't find one after trying, stop. If you find one, you haven't saturated; you've found a group you didn't know you had. Go back to step one.
Because the alternative is worse. Teams that believe they need "statistically significant" qualitative research either never start, or run forty interviews with no groups and no coding and drown. The shortcut gives a small team a way to plan, budget and stop with a straight face — as long as they plan by groups, stop by saturation, and hunt for the breaker.
Quote the number if you like. Quote the three steps with it, or you're quoting something I didn't say.
The full method, with the tracker: So They Lied to You / «Так тобі й збрехали»
Nina Somkina · Founder, Institute for Human Reasoning
Institute for Human Reasoning is a research and education organisation strengthening human reasoning for consequential decisions in a technology-shaped world.
Seventy years of research on evidence and decisions, in six findings that still embarrass every company calling itself data-driven.
ReadEvery company will say yes. The only measurement that means anything is where AI enters a decision — and whether anyone checks it when it disagrees with the boss.
Read