Both user interviews and surveys are essential UX research tools, but they answer fundamentally different questions. Choosing the wrong one for your current question wastes time and produces misleading data. Choosing the right one, or the right combination, can be the difference between a product that resonates and one that misses.
What User Interviews Give You
User interviews are one-on-one conversations, typically 45 to 60 minutes, that explore a participant's goals, behaviours, mental models, and experiences in depth. They are qualitative: the goal is not to count but to understand.
Interviews are exceptional at surfacing the why. Why does a user use a workaround instead of the intended feature? Why do they abandon a flow midway? Why does a certain task feel stressful? These motivations, emotional states, and contextual factors are almost impossible to capture in a survey.
The limitation of interviews is scale. You typically speak to 5–15 participants. These insights are rich but cannot be generalised with statistical confidence to a larger population without further validation.
What Surveys Give You
Surveys collect self-reported data from many respondents simultaneously, enabling statistical analysis at scale. They are quantitative: they tell you how many users feel a certain way, how often they perform a behaviour, or how satisfied they are on a measurable scale.
Surveys are best for validating patterns, measuring sentiment across your user base, and tracking changes over time (for example, running the same NPS survey quarterly to track satisfaction). They can reach hundreds or thousands of users efficiently.
The limitation of surveys is depth. Users can only answer the questions you ask, and they often do not know why they behave the way they do. "I find the product easy to use" is useful data, but it cannot tell you which specific interaction is causing confusion, or why a user formed that impression.
When to Choose Interviews
Choose user interviews when you are in discovery mode, when you do not yet know what questions to ask in a survey. When you need to understand user goals, mental models, or the context around behaviour. When you are designing something new and need to understand who you are designing for. When you have a hypothesis about user behaviour and need to explore whether it is true.
Five to eight well-recruited participants who match your target user profile will reveal the most important insights about your product's problem space.
When to Choose Surveys
Choose surveys when you need to validate or quantify patterns already identified through qualitative research. When you want to know the scale of a problem ("how many users experience this?"). When you need to track a metric over time. When your user base is too large or geographically dispersed to interview efficiently.
Surveys are also useful for post-launch satisfaction measurement, feature prioritisation voting, and demographic profiling of your user base.
Using Both Together
The most powerful research programmes combine both methods sequentially. Interviews first: run five to ten in-depth interviews to understand user behaviour and surface the most important themes. Then survey to validate: turn those themes into survey questions to measure their prevalence across your broader user base. This gives you both the rich understanding that interviews provide and the statistical confidence that surveys enable.
Common mistake: Running a survey before any interviews means you are asking questions you invented, not questions grounded in actual user behaviour. You will measure things that may not matter and miss things that do.
Key Takeaways
- Interviews reveal why users behave as they do; surveys measure how many and how often.
- Use interviews in discovery, when you need to understand, not count.
- Use surveys to validate at scale patterns found in qualitative research.
- Running interviews before surveys ensures your survey questions are grounded in real user behaviour.
- Five to eight interviews is enough for most discovery questions; survey sample sizes should match statistical confidence needs.
