By the time I sat down for my first UX user interview, interviewing wasn’t the scary part. I’d spent seven years as a technical recruiter, and I’d talked to more strangers over video and phone than I could count. Hitting record and asking someone to open up didn’t rattle me. The problem showed up the moment I started writing my discussion guide.
I kept catching myself writing the same move I make on every recruiting call: paraphrase what someone just told me, then hand it back as a question. “Based on what you’ve told me, would you say that…” “Is it safe to assume that…” Those phrases keep a candidate talking. In a UX interview, they do the opposite. They tell the participant what answer you’re hoping for, and most people, without meaning to, will hand it right back to you.
That’s the moment recruiting stopped feeling like an advantage and started feeling like a liability. As a recruiter, I’m rarely the technical subject matter expert on a call. My job is to pull as much signal as possible out of someone in thirty minutes, so I paraphrase constantly, both to confirm I understood and to keep the conversation moving. It works great for screening candidates. It is close to the textbook definition of a leading question, and leading questions are exactly what you don’t want in a user interview.
The stakes felt personal too. The topic I picked for my final project, and the survey work that grew out of it, was a problem I’d lived with myself, alongside a peer community I’m part of. I’d spent years as a passive and active observer of it. That’s exactly the setup where a leading question does the most damage. I already had a theory, and it would have been easy to write questions that just confirmed it.
Unlearning the paraphrase habit took real effort. After seven years as a technical recruiter, it was muscle memory. But three things I’d built over those seven years turned out to transfer almost directly into running good user interviews: listening more than I talked, getting comfortable with silence, and knowing how to bring a rambling conversation back on topic without making the other person feel like they’d done something wrong. What follows is what happened when I held those recruiter instincts up against the actual UX research literature and checked which ones held up.
Writing user interview questions that don’t lead the witness
The fix wasn’t better instincts. It was a checklist. Nielsen Norman Group researcher Amy Schade has written the clearest definition I’ve found: a leading question includes or implies its own answer, which makes it hard for a participant to say anything else. Schade names four specific traps, and I’d been walking straight into two of them on every recruiting call: paraphrasing what someone just said back to them, and suggesting an answer inside the question itself. The other two are naming an interface element before the participant does, and naming their emotional state for them before they’ve named it.
Schade’s article includes rephrasing pairs I now keep taped next to my monitor:
| Leading (avoid) | Neutral (use instead) |
|---|---|
| “I saw you were having difficulty with the navigation. What happened?" | "What was easy or difficult about getting to the content you wanted?" |
| "How well would this save you time during your workday?" | "How might this affect your efficiency, if at all?" |
| "The related links on the side of the page here, where would those lead?" | "This area on the side of the page here. What is that?” |
Good user interview questions describe, they don’t diagnose. “What was easy or difficult about X” leaves room for “nothing, it was fine.” “I saw you were having difficulty” doesn’t. That distinction matters more than it looks. A persona built on a handful of leading answers is a persona built on what the researcher already believed, not on what users actually think, and it’s a lot harder to unwind six months later than it is to fix a script before session one.
The four-second wall, and why I count to seven
Silence isn’t dead air. It’s where the real answer lives. This is the one recruiter habit that transferred over without any adjustment. Candidates go quiet when they think they’ve fully answered a question, and if I jump in too fast, I cut them off before the actual insight arrives. NN/g researcher Kate Kaplan put a number on why that jump feels so urgent: citing research on conversational discomfort, she notes that in the US, silence starts to feel uncomfortable after about four seconds. Kaplan’s own rule of thumb is to silently count to seven before speaking again. The first four seconds is where the discomfort kicks in. The extra three is what actually gives someone room to keep going, past their first, tidiest answer, into the messier and more useful one underneath it.
I use a version of this in every session now. I tell participants upfront that if I go quiet, it’s because I’m taking notes, not because I’ve checked out. That one sentence does two things. It gives me permission to sit inside an uncomfortable pause instead of filling it, and it heads off a related mistake researchers flag often: visibly scribbling mid-answer signals to a participant which parts of their response you found interesting, which is its own quiet form of leading. Kaplan’s number isn’t universal, either. The same research notes Japanese speakers tolerate roughly double the silence English speakers do, so “count to seven” is a US baseline, not a law of nature. Know your participant before you assume your own comfort with quiet is theirs.
Giving the conversation a shape
A good user interview script isn’t a list of questions. It’s a shape. On recruiting calls, I always tell the candidate how the conversation will run before we start: logistics first, then their background, then my questions, then time for theirs. That turns out to be close to the standard structure researchers use. NN/g and independent practitioners like Stéphanie Walter both describe a four-part shape: an introduction that states what you’re trying to learn and re-confirms consent to record, warm-up questions to build rapport before the real material starts, a body organized around three to five topic areas, and a wrap-up that asks if there’s anything the participant wants to add. Thirty to forty-five minutes is enough time to run all four without wearing anyone out.
Telling someone the shape of the conversation upfront does the same thing for a UX participant that it does for a candidate. It tells them when it’s their turn to talk, which cuts down on interruptions and nerves. What I never had a name for, until I read the research, was what to do with the good stuff that comes up off-script. My recruiter instinct was to just call it improvising. The actual techniques are called Five Whys and laddering, and they’re structured versions of the same move. Five Whys drills straight down on one thread by repeatedly asking why. Laddering branches sideways and upward toward what the participant actually values. IDEO’s field guide adds a third option I use constantly now, called “Show Me”: instead of asking someone to describe how they use something, ask them to pull it up and walk you through it live. It’s the same instinct behind watching someone actually work through an onboarding flow instead of asking them to describe it from memory afterward. People are bad narrators of their own behavior. They’re much better demonstrators of it.
A quick aside, since this keeps coming up with every research team I talk to: User Interviews’ 2025 State of User Research report found 80% of researchers now use AI tools in their work, up 24 points from the year before. Maze’s 2026 research report puts AI adoption at 69%, mostly for transcription and synthesis. Both are vendor surveys selling research tooling, so treat the exact percentages as directional, not gospel. But the trend lines agree, and so does the caveat buried in Maze’s own data: AI can summarize a transcript, but it can’t pick up on hesitation or decide which insight actually matters. That’s still the moderator’s job. Count-to-seven doesn’t automate.
Reflective Coda
Seven years of recruiting calls didn’t make me a good UX researcher. They gave me raw material: comfort with strangers, a decent ear for when someone’s being polite instead of honest, and a bad habit that needed naming before I could break it. The named frameworks did the rest. I don’t write a leading question anymore without catching it, because I know what a leading question is, structurally, not just vaguely. I don’t fill a silence anymore, because I know the four seconds is the discomfort talking, not the participant being finished.
If you’re coming into UX research from another people-facing job, sales, teaching, therapy, customer support, don’t throw out your instincts. Audit them. Some of what feels like a natural conversational skill is exactly what a rigorous interview needs. Some of it is a bias with good manners. The only way to tell the difference is to hold your habits up against the actual research and see which ones survive.
These days, when I sit down for a user interview, the nerves are still gone, same as they were on recruiting calls. What’s different is what I do with the quiet. I used to fill it. Now I count to seven and let someone tell me the truth.
