How many interviews are enough for a generative study?

Most generative studies run 8-15 interviews per segment. Saturation — the point where new interviews stop producing new codes — is a better signal than a fixed count. If interview 12 still surfaces new themes, keep going. The Nielsen rule of 5 applies to usability testing, not generative work.

What is the difference between a listening session and a regular interview?

Indi Young's distinction: a listening session has no agenda beyond a topic; the participant guides while the researcher follows their thinking. A regular interview has specific questions to answer. Listening sessions surface deeper mental-model insight; interviews surface targeted answers. Senior UXRs use both, matched to the question.

When should I use a diary study instead of interviews?

Diary studies fit when behavior unfolds across days or weeks, when participants would forget the context by interview time, or when you need to see routines in their actual environment. They are expensive and harder to recruit for, so save them for questions a one-shot interview cannot answer.

How do I run contextual inquiry without being intrusive?

Observe first, ask second; let the participant explain in their own words; avoid interrupting flow. Beyer and Holtzblatt's apprenticeship framing — 'I am the apprentice, you are the master' — sets the right power dynamic. Ask clarifying questions during natural pauses.

What is Jobs to Be Done and when should I use it?

JTBD frames findings around the progress a participant is trying to make rather than features they request. Use it when the deliverable is a strategy artifact (opportunity map, jobs framework, switch analysis) rather than a usability fix. Tony Ulwick's outcome-driven JTBD structures jobs as situation-motivation-outcome; Bob Moesta's switch-interview JTBD focuses on the forces that drove a switch.

How do I synthesize an interview corpus without an expensive tool?

A spreadsheet is sufficient: one row per coded moment with columns for participant ID, timestamp, quote, code, theme. Sort by code, eyeball clusters, draft theme labels, restructure rows by theme. The discipline matters more than the tool. Dovetail or Reduct make it faster, but a senior UXR with a spreadsheet outperforms a junior with the best tooling.

How do I avoid leading questions during a discovery interview?

Three habits: (1) ask about past behavior, not future intent ('tell me about the last time you did X' beats 'would you use a feature that did X'); (2) use 'tell me about' rather than 'do you'; (3) tolerate silence — the second answer after a pause is usually more honest. Erika Hall and Indi Young both train this discipline.

How do I report generative findings to a skeptical executive audience?

Lead with implications, support with evidence, link to the corpus. Structure: top-line insight, strategic implication, 3-5 representative quotes with participant IDs, and a link to the repository. Executives respond to defensibility — drilling from claim to quote builds trust faster than slide polish.

Can I run generative research with a tight timeline?

Yes — Erika Hall's Just Enough Research is built around this case. Compressed version: 5-8 interviews instead of 15; lighter synthesis; insights as a memo rather than a deck. Skip nothing methodologically; just do less of it. The trap is skipping synthesis — that is where the value is.

UX Researcher Hub

Generative Research and Discovery for UX Researchers (2026)

By Blake Crosley · Last verified 2026-04-30

In short

Generative research surfaces what people are trying to do, why, and the constraints that shape behavior — before a product team has decided what to build. The toolkit: 1:1 depth interviews (Indi Young's listening sessions, Erika Hall's pragmatic interviews), diary studies, contextual inquiry, field ethnography, and Jobs to Be Done framing. Senior UXRs use generative methods when the team's questions are about goals, motivations, and context — not whether a known design works. Synthesis turns transcripts into a defensible insight set that anchors strategy.

Key takeaways

Generative research answers 'what are people trying to do, why, in what context' — not 'does this design work.' Erika Hall's Just Enough Research is the canonical pragmatic guide.
1:1 depth interviews are the workhorse method. Indi Young's listening sessions train the interviewer to follow the participant's mental model rather than steer toward features.
Diary studies capture behavior across time and context that a one-shot interview cannot — routines, frustrations, and triggers participants forget by interview time.
Contextual inquiry and field ethnography put the researcher in the participant's environment. Watching someone do the work reveals workarounds no interview surfaces.
Jobs to Be Done framing reframes findings around the progress a participant is trying to make. Tony Ulwick's outcome-driven and Bob Moesta's switch-interview JTBD are the two canonical schools.
Discovery structure follows a story-spine (context → trigger → behavior → outcome → reflection) plus the 5 Whys for motivation. The structure keeps the interviewer from leading.
Synthesis is the hard part. A study produces 8-20 hours of transcript that must be coded, clustered, and distilled into defensible insights. Without rigorous synthesis, generative research becomes anecdote farming.

When generative research is the right method

Generative research is the right method when the team's questions are about goals, motivations, and context of use — not whether a known design works. The canonical framework comes from Christian Rohrer's UX methods landscape (Nielsen Norman Group): generative work sits in the qualitative + attitudinal-and-behavioral quadrant, used early when the problem space is uncertain.

Use generative methods when:

The team is exploring a new problem space. The PM has a hunch, not a design. The output is a problem definition, user goals, and a sense of who matters.
You are entering a new market or audience. The team's mental model is built from a different segment; generative work surfaces what is genuinely different.
Behavior is unexplained. Quantitative data shows a pattern (drop-off, low adoption) and the team has no story for why.
JTBD or strategy work is the deliverable. The output is a Jobs framework, opportunity map, or persona refresh — not a usability fix.

Do not use generative methods when the team has a specific design and wants to know whether users can complete a task (that is evaluative usability work), when the team needs a number (that is quantitative survey work), or when the team has already decided and is looking for confirmation. Generative research surfaces inconvenient truths; running it as a confirmation exercise is malpractice.

Erika Hall's Just Enough Research (muleshq.com) is the canonical pragmatic guide. Tomer Sharon's Validating Product Ideas (tomersharon.com) covers the same decision in a product-discovery frame. Nielsen Norman Group's methods guide (nngroup.com/articles/which-ux-research-methods) and qualitative-vs-quantitative reference (nngroup.com/articles/qualitative-vs-quantitative-research) are the canonical academic references.

Interview design that produces signal

A generative interview is a structured conversation in which the researcher follows the participant's mental model rather than a feature checklist. The senior bar is interviewer discipline: you ask open questions, you tolerate silence, and you do not lead.

The canonical structure is a story-spine:

Context. Who are you, what does a typical week look like, where does this work fit in your life?
Trigger. When did you last do the thing we're researching? Walk me through what was happening that day.
Behavior. What did you actually do — step by step? What tools did you reach for, who did you talk to?
Outcome. How did it turn out? What worked, what didn't?
Reflection. Looking back, what was the hardest part? What surprised you?

Inside the spine, the 5 Whys drives toward motivation. When a participant says 'I switched tools,' the researcher follows with 'why did you switch then?' — and again, until the underlying motivation surfaces. Surface answers are usually rationalizations; the third or fourth 'why' reaches the actual driver.

Indi Young's listening-session methodology (indiyoung.com, Practical Empathy) is the most rigorous training for this discipline: a listening session is the participant teaching the researcher; the researcher's job is to be a careful student, not a question-asker. Erika Hall's Just Enough Research covers the pragmatic version.

Common failures:

Leading questions. 'Don't you find the current process frustrating?' is a request for confirmation. Replace with 'tell me about the last time you did this.'
Hypothetical questions. 'Would you use a feature that did X?' produces fiction. Replace with 'when was the last time you tried to do X?'
Feature-anchored questions. 'What do you think of [our feature]?' anchors the participant in the team's solution rather than their problem.
Interviewer talking. Junior interviewers fill silence; senior interviewers wait. The participant's third sentence after a long pause is usually the real answer.

Jared Spool's writing on interviewing technique (uie.com) is the canonical practitioner reference for the silence-and-follow-up discipline.

Synthesis: from raw notes to insight

A generative study produces 8-20 hours of transcript. The synthesis problem is to turn that corpus into a defensible insight set — typically 5-10 insights the product team can act on without re-reading the source material.

The canonical workflow:

Capture. Every interview produces a transcript plus a research-note artifact — quotes, behaviors, surprises tagged in real time. The corpus is the source of truth.
Open coding. Tag every relevant moment with a short descriptive code (e.g., 'workaround for slow approval'). The researcher generates codes rather than applying a predefined taxonomy — that is what lets unexpected patterns surface.
Affinity clustering. Group codes into themes via affinity diagram (sticky notes, or FigJam / Mural / Dovetail). Themes emerge from proximity, not predefined categories.
Insight statements. Each cluster condenses to a finding plus evidence plus strategic implication. 'Users in segment X work around constraint Y by doing Z, because their underlying goal is W' — supported by quotes from N participants.
JTBD framing (when applicable). Restructure insights around the progress a participant is trying to make. Tony Ulwick's outcome-driven JTBD structures jobs as 'when [situation], I want to [motivation], so I can [outcome]'; Bob Moesta's switch-interview JTBD focuses on the forces (push, pull, anxiety, habit) driving a switch.
Repository deposit. Insights, raw clips, and tags go into the team's research repository so future researchers re-find evidence rather than re-running the study.

The tooling has converged in 2026: Dovetail, Reduct, Condens, Marvin for transcription + tagging; FigJam and Mural for affinity work; spreadsheets for teams without a repo budget. Discipline matters more than tooling.

Common synthesis failures: cherry-picking quotes that confirm a pre-held belief; coding too coarsely so no pattern emerges; skipping open coding; producing insights without evidence trails — 'users want X' with no link to the corpus is opinion, not research.

Common pitfalls in generative work

Generative research is the easiest UXR work to do badly because the methodological floor is so low — anyone can run an interview. The senior bar is the discipline to do it well.

Recruiting the wrong participants. Convenience samples (coworkers, Twitter friends) produce confirmatory anecdote, not signal. Screen aggressively up front.
Confusing generative with evaluative. A team books 'discovery interviews' but spends 40 of 60 minutes showing wireframes. That is evaluative work disguised as discovery. Pick one; do not blend.
Anecdote farming. Pulling memorable quotes without rigorous synthesis. The insights are unfalsifiable — no way to know whether a quote represents a pattern or a one-off.
Over-claiming sample size. Insights from 8 interviews describe 8 people in depth, not the user base in aggregate. Erika Hall's framing: qualitative tells you what; quantitative tells you how many.
Skipping synthesis. Run interviews, post quotes in Slack, call it research. That is theater. The synthesis is where the value is.
Researcher bias in coding. The researcher who ran the interviews codes the corpus and finds what they expected. Mitigation: a second coder or peer review of the insight set.
Letting methodology drive the question. 'We need to do a diary study' before knowing what question is being asked. Method follows question.
No theory of impact. Study finishes; nothing changes. Design studies with the decision in mind: which roadmap question does this inform, what would have to be true for the team to act.

Frequently asked questions

How many interviews are enough for a generative study?: Most generative studies run 8-15 interviews per segment. Saturation — the point where new interviews stop producing new codes — is a better signal than a fixed count. If interview 12 still surfaces new themes, keep going. The Nielsen rule of 5 applies to usability testing, not generative work.
What is the difference between a listening session and a regular interview?: Indi Young's distinction: a listening session has no agenda beyond a topic; the participant guides while the researcher follows their thinking. A regular interview has specific questions to answer. Listening sessions surface deeper mental-model insight; interviews surface targeted answers. Senior UXRs use both, matched to the question.
When should I use a diary study instead of interviews?: Diary studies fit when behavior unfolds across days or weeks, when participants would forget the context by interview time, or when you need to see routines in their actual environment. They are expensive and harder to recruit for, so save them for questions a one-shot interview cannot answer.
How do I run contextual inquiry without being intrusive?: Observe first, ask second; let the participant explain in their own words; avoid interrupting flow. Beyer and Holtzblatt's apprenticeship framing — 'I am the apprentice, you are the master' — sets the right power dynamic. Ask clarifying questions during natural pauses.
What is Jobs to Be Done and when should I use it?: JTBD frames findings around the progress a participant is trying to make rather than features they request. Use it when the deliverable is a strategy artifact (opportunity map, jobs framework, switch analysis) rather than a usability fix. Tony Ulwick's outcome-driven JTBD structures jobs as situation-motivation-outcome; Bob Moesta's switch-interview JTBD focuses on the forces that drove a switch.
How do I synthesize an interview corpus without an expensive tool?: A spreadsheet is sufficient: one row per coded moment with columns for participant ID, timestamp, quote, code, theme. Sort by code, eyeball clusters, draft theme labels, restructure rows by theme. The discipline matters more than the tool. Dovetail or Reduct make it faster, but a senior UXR with a spreadsheet outperforms a junior with the best tooling.
How do I avoid leading questions during a discovery interview?: Three habits: (1) ask about past behavior, not future intent ('tell me about the last time you did X' beats 'would you use a feature that did X'); (2) use 'tell me about' rather than 'do you'; (3) tolerate silence — the second answer after a pause is usually more honest. Erika Hall and Indi Young both train this discipline.
How do I report generative findings to a skeptical executive audience?: Lead with implications, support with evidence, link to the corpus. Structure: top-line insight, strategic implication, 3-5 representative quotes with participant IDs, and a link to the repository. Executives respond to defensibility — drilling from claim to quote builds trust faster than slide polish.
Can I run generative research with a tight timeline?: Yes — Erika Hall's Just Enough Research is built around this case. Compressed version: 5-8 interviews instead of 15; lighter synthesis; insights as a memo rather than a deck. Skip nothing methodologically; just do less of it. The trap is skipping synthesis — that is where the value is.

Sources

About the author. Blake Crosley founded ResumeGeni and writes about UX research, hiring technology, and ATS optimization. More writing at blakecrosley.com.