Semi Structured Interviews In Qualitative Research

The Guide Nobody Asked For

Semi-structured interviews are the most used tool in qualitative research for good reason. You walk in with a guide of topics and open-ended questions, then let the conversation go wherever the participant takes it. It sounds simple. It is not. The actual work happens in the gap between your prep work and what the participant says. I spent three years doing this for a thesis that ended up needing about forty interviews before I felt like I had enough signal. The first twelve were a mess because I kept treating the interview guide like a checklist instead of a map.

What You Actually Do With Semi Structured Interviews In Qualitative Research

Start by writing a discussion guide. Not a questionnaire. A guide. It should have three to five broad themes with two or three probe questions under each theme. The probes are where the quality of your data lives. "Can you tell me more about that?" and "What did that feel like for you?" count as probes. They are underused and they should not be. Keep the document to about two pages maximum. If it is longer, you will read from it and the conversation dies. I learned this the hard way during a project studying workplace burnout. My guide was four pages long because I was nervous. The first six participants gave me monosyllabic answers and I had no idea why until I played back the recordings and realized I was reading verbatim instead of listening. The process itself is straightforward on paper. You recruit people who fit your sampling criteria, schedule forty-five to ninety minute sessions depending on your topic, record with permission, and transcribe. The recording is optional but strongly recommended because you will miss details when you are trying to maintain rapport. Write field notes afterward while the interaction is fresh. Those notes matter more than you think.

One specific problem I ran into constantly involved participants who were either too brief or too verbose. There is a narrow middle band where you get usable data without exhausting yourself. When someone gives only one sentence answers, you pivot to silence and wait. Not awkward silence. Just neutral silence. Most people fill it. When someone rambles for ten minutes about something irrelevant, you use a gentle redirect like "That is interesting, I want to come back to what you said earlier about X." You do not apologize for doing this. It is your job to steer. The transcription step is where people cut corners and regret it later. Automated tools like Otter or Descript have gotten decent but they still butcher quotes with medical terminology or technical jargon. If your study involves any specialized vocabulary, budget extra time for manual verification. A thirty minute interview typically takes forty five minutes to transcribe accurately by hand, give or take depending on overlap and accent clarity. Analysis comes after all interviews are done, not during. This is counter-intuitive for beginners who want to start coding after the first conversation. You can start noting patterns as you go, which is called iterative analysis, but real thematic work happens when you have seen enough variation to know what is actually significant versus what is anecdotal noise. I usually aim for somewhere between fifteen and twenty five interviews before thematic saturation becomes visible. Your topic determines the number. Rare populations need fewer. Common ones need more.

The biggest pitfall people hit is confirmation bias disguised as open inquiry. You think you are being neutral but your follow-up questions are subtly pushing toward what you already expect. The fix is simple in theory and hard in practice. Have a colleague review your interview guide before you start. Better yet, pilot it with two people who do not fit your target demographic. If your questions do not work on someone outside your sample, they will not work inside it either. Another thing nobody warns you about is the relationship between interview length and data quality. Longer is not automatically better. Beyond about seventy five minutes, participant fatigue sets in and the depth drops even though the word count goes up. You end up with padding. I started capping sessions at sixty five minutes and my data actually improved because the conversation stayed sharp.

Setting Up Your Interview Guide

Write your opening script so you do not have to think about it during the call. It should cover consent, recording, confidentiality, and an invitation to ask questions. Read it the same way every time. Consistency matters less than people admit but the opening sets the tone for the entire session. Your main questions should be open-ended and neutral. Avoid questions that contain their own answer. "Don't you think remote work is harder than office work?" is a terrible question. "How would you describe the difference between working remotely and working in an office?" is functional. The difference between those two questions is the difference between collecting opinions and collecting data. Probe questions live in parentheses under each main question. You use them when the participant gives a surface-level answer or when a new direction emerges that relates to your theme. Having them written down saves you from freezing mid-conversation trying to remember what you wanted to ask next.

Closing questions are important and routinely skipped. Always end with something like "Is there anything we have not talked about that you think is important?" and "Do you know anyone else who might have useful perspective on this?" The second part is snowball referral and it has saved my recruitment multiple times.

Common Failures and How to Avoid Them

Not all semi-structured interviews are equal and some designs will fail regardless of your skill. If you are studying sensitive topics like trauma or illegal activity, this method alone may not get you honest data. Participants will self-censor in ways that are visible during transcription but invisible during the interview. Pair it with anonymous surveys or use a different method entirely. Being honest about method limitations is part of doing the work correctly. Another scenario where this breaks down is when your population has very low literacy or limited comfort with extended conversation. I worked on a project involving elderly participants in rural areas where the standard interview format produced mostly frustrated silence. Switching to shorter sessions with more visual prompts and breaks made the difference between usable and unusable data. If you need hard numbers to justify sample size, there is no universal formula. Grounded theory studies often stop around twenty interviews. Phenomenological studies might need fifteen. Generic thematic analysis projects I have overseen usually land between twenty and thirty before the research team agrees nothing new is emerging. Track your saturation explicitly in your methodology section rather than hiding behind vague language.

Documentation matters for credibility. Keep a reflexive journal alongside your transcriptions. Note what surprised you, what frustrated you, and where your own assumptions may have shaped the conversation. Review entries between interviews. This habit alone prevented me from chasing a dead-end theme for six weeks on a project that ended up pivoting in a much more productive direction after interview nine. The method works well enough to be the default choice for most qualitative projects. It does not work well when you treat it as a checkbox exercise or expect it to replace careful thinking about sampling, analysis, and your own positionality. Spend time on the guide. Record everything. Transcribe carefully. Let the data talk back.