Semi-structured interview: fixed core questions, free follow-ups

A semi-structured interview is a one-to-one conversation where you ask every person the same short list of core questions, then follow up on whatever their answers bring up. The fixed core lets you compare people. The free follow-ups let you hear things you did not think to ask.

Most of the skill is deciding which parts stay fixed and which parts can move. Get that split right and the analysis almost organizes itself. Get it wrong and you end up with eight pleasant conversations that cannot be compared.

1. Choose semi-structured when you know the topics but not the answers

Interviews are usually sorted by how much is decided in advance. The UK Statistics Authority describes a structured interview as one where questions are asked in a set order with no variation between participants, a semi-structured interview as one built on a set of key questions chosen from the research aims (plus probing follow-ups), and an unstructured interview as informal and flexible, with no topic guide.

Yan Zhang and Barbara Wildemuth's chapter on unstructured interviews fills in the rest. Structured interviews are close to surveys read aloud. In a semi-structured one, the interviewer has room to change the order and add questions based on what the person says.

The first two weaknesses follow from the format itself. For the third, Zhang and Wildemuth note that unstructured interviews can run longer, are more open to the interviewer's influence, and take a great deal of effort to analyze because every session asks different questions. Nielsen Norman Group adds a practical point: when the same few open questions are asked of each person, overlaps show up sooner, so you reach the point of hearing nothing new with fewer interviews.

Market researchers call the semi-structured format a depth interview, as NN/g notes, so in-depth interviews covers the same conversation from the study-planning side.

For most founders and PMs, semi-structured is the default. You know you want to learn how procurement works or why people churned, but you do not know what they will say. If you already know the likely answers and need to count them, run a survey instead (interviews versus surveys). If you are hiring, a fully structured interview is the fairer tool.

2. Write five to ten core questions and mark the ones you always ask

Melissa DeJonckheere and Lisa Vaughn's 2019 review of semistructured interviewing describes the guide as a short list of guiding questions, with five to 10 core questions that relate directly to what you want to know. Around them sit three other kinds:

1. A grand tour question: a broad opener about something the person knows a lot about.
2. Planned follow-ups: specific questions about parts of a core question.
3. Unplanned follow-ups: questions that come up during the interview from what the person said.

They recommend that every question be open, neutral and clear, and that the interview start with an easy question that sets context. William Adams, in his chapter for the Handbook of Practical Program Evaluation, calls about one hour a reasonable maximum length for these interviews, to limit fatigue on both sides.

Mark each line in your guide as one of two kinds. Core questions get asked of everyone, in roughly the same words. Planned follow-ups get asked only if the person has not already covered them, which is how NN/g suggests using them.

Here is a made-up example. Priya is a PM at a company that makes maintenance software for apartment buildings. She wants to learn how property managers at buildings with 50 to 200 units handle a repair request from first report to closed ticket. Her core questions:

1. Tell me about your role and the buildings you look after. (grand tour)
2. Think of the last repair request that came in. How did you hear about it?
3. What happened between that and someone showing up to fix it?
4. How did you know the job was actually done?
5. How did that request compare with a typical week?
6. What would you change about how requests are handled, and why?

Each gets two planned follow-ups, such as "Who else touched it?" under question 3. Six core questions with room to probe fits a 45-minute call; 15 would not.

Short version: fewer core questions give you more time to follow up, and the tradeoff is less you can compare. Five to 10 keeps both. For building the full guide (introduction, timings, closing), use how to write a discussion guide; for wording individual questions, see qualitative research interview questions.

3. Probe from a short list instead of inventing questions on the spot

Probes are the small moves that get someone to say more. DeJonckheere and Vaughn list six, adapted from the anthropologist H. Russell Bernard:

The paper labels the last one "leading". It means leading the person to explain their reasoning, not putting words in their mouth, so call it something else on your own sheet.

Print the table and keep it next to the guide. GOV.UK's service manual offers the same kind of simple prompts, such as "you said... when/why/who was that?" and "in what way...?" They work because they carry no opinion of their own.

Wait time is the hardest probe. Five seconds of silence on a video call feels like a minute. (It is still five seconds, and the person usually fills it with the useful part.)

4. Pilot the guide, then version it

DeJonckheere and Vaughn say to pilot test the interview once the guiding questions are written. NN/g's guide on interview guides suggests at least one pilot participant, to catch missing questions, unclear wording, bad ordering and timing problems. It also says updating the guide during the study is acceptable.

The catch is the core. If Priya rewrites question 4 after her fourth interview, interviews one to four and five to eight answered different questions. Keep a version number and a dated line on the guide for every change, so you know which answers can sit side by side.

On how many interviews, DeJonckheere and Vaughn note there is no power calculation and the honest answer is "it depends", while also observing that a meaningful project can involve as few as 8 to 12 participants. NN/g suggests starting with a small representative sample of five or six and analyzing as you go. How many customer interviews you need covers when to stop.

5. Analyze question by question first, then look at what you did not plan

The fixed core is what makes analysis manageable, so use it. Nicola Gale and colleagues describe the Framework Method, which they say is most commonly used for thematic analysis of semi-structured interview transcripts. Summarized data goes into a matrix: one row per person (a "case") and one column per code. Its seven stages are transcription, familiarization, coding, developing a working framework, applying it, charting into the matrix and interpreting.

They also warn that the method cannot handle highly heterogeneous data. The interviews must cover similar topics. That is your core questions doing their job.

For a small product team, a lighter version works:

1. Record and transcribe. DeJonckheere and Vaughn recommend audio recording so you can concentrate on the person, and warn that transcription is time-consuming. Ask first (recording consent).
2. Build a spreadsheet. One row per person, one column per core question, and one extra column called "unplanned" for anything outside the guide.
3. Fill each cell with what happened, in a sentence or two, with a quote where it helps.
4. Read down each column. This is where patterns across people show up.
5. Read the unplanned column last. Surprises live here.

Back to the made-up example. After eight interviews, Priya reads down the question 4 column and sees that five of the eight managers only learn a job is done when the tenant stops complaining. Three mention, unprompted, that they keep a paper log in the leasing office. Neither finding needed a percentage, and "five of eight" says exactly what it is.

NN/g's thematic analysis guide defines a theme as something that emerges when related findings appear multiple times across participants, and warns that analysis is often done superficially, just skimming topics. A column summary is a topic. A theme explains why the column looks the way it does. For the step from notes to conclusions, see how to analyze customer interviews and affinity mapping.

Your next step

Write your five to 10 core questions on one page, add two planned follow-ups under each, and run the guide once with a colleague who plays a participant. If you need real participants outside your customer list, Instant Expert can find people matching a description such as "property managers at buildings with 50 to 200 units" and send each a paid invitation for a short call, and you pay only if someone books. The multifamily property operations experts and user research experts directory pages are places to start.