Generative Research: How to Uncover User Needs You Didn't Know Existed

Generative Research: How to Uncover User Needs You Didn't Know Existed

Generative research is open-ended investigation designed to discover user behaviors, motivations, and unmet needs before you know what problem to solve. Unlike evaluative research — which tests whether something works — generative research asks a more fundamental question: what should we build, and for whom?

Teams that skip generative research often build the wrong thing brilliantly. The product ships, the engineering is clean, and nobody uses it — because the underlying need was never truly understood. Generative research prevents that failure mode.

---

What Is Generative Research?

Generative research (also called discovery research or exploratory research) produces insights, hypotheses, and opportunity areas rather than validating existing assumptions. The output isn't a yes/no answer — it's a rich understanding of how people think, what they struggle with, and what they wish existed.

Common generative research questions sound like:

• "How do product managers currently share research insights with engineering?"
• "What does the first 30 days feel like for a new hire in a remote-first company?"
• "How do small business owners decide when to hire their first employee?"

These questions don't presuppose a solution. They open space for genuine discovery.

---

Generative vs. Evaluative Research

Both are essential. Generative research gives you the right problem to solve; evaluative research confirms you've solved it well.

---

When to Run Generative Research

Run generative research when:

You're entering a new market or user segment. You don't know enough about how people live and work in this space to make informed product decisions.

You've hit a growth plateau. Users are adopting your product but not expanding or retaining. Generative research reveals what's missing or misaligned with their actual workflows.

You're planning a major feature or pivot. Before investing in a new direction, understanding user context prevents expensive course corrections.

Your team is working from assumptions. When everyone has opinions but nobody has talked to users recently, generative research regrounds the team in reality.

Research shows teams that conduct regular discovery research ship features users want at 2–3x the rate of teams that skip it. With AI-powered interview platforms like Koji, running generative research at scale takes hours — not weeks.

---

The Best Methods for Generative Research

In-Depth Interviews

One-on-one conversations are the gold standard for generative research. They let you probe deeply, follow unexpected threads, and understand not just what people do but why. Open-ended questions like "Walk me through the last time you..." unlock rich narrative data.

Traditional interviews require scheduling, moderation, and manual analysis. Platforms like Koji automate the moderation and analysis, letting you run 50 interviews in the time it used to take to do 5. Koji's AI interviewer asks intelligent follow-up questions in real time, probing naturally the way a skilled human researcher would — while its six structured question types (open-ended, scale, single-choice, multiple-choice, ranking, and yes/no) let you collect both qualitative depth and quantitative structure in the same session.

Ethnographic Research

Ethnographic research involves observing users in their natural environment. You're watching, not asking — seeing the workarounds, the sticky notes on monitors, the frustrations people don't even notice anymore because they've normalized them. This method surfaces tacit knowledge: the things people can't articulate in an interview because they've stopped seeing them as problems.

Ethnographic research is most valuable when context and environment are central to the behavior you're studying. It's resource-intensive but often produces the most surprising and actionable discoveries.

Diary Studies

Diary studies ask participants to record thoughts, behaviors, or experiences over time — typically days or weeks. They capture longitudinal patterns that point-in-time interviews miss. Koji's async interview format supports diary-style research naturally: send participants periodic check-in prompts to capture in-the-moment reactions without the overhead of a formal diary protocol.

Contextual Inquiry

A hybrid of observation and interview, contextual inquiry involves shadowing a user as they work and asking questions in the moment. "What are you doing there?" "Why did you switch tools?" The context makes answers more specific and honest than retrospective recall.

Expert Interviews

When you're entering an unfamiliar domain, interviews with subject matter experts — practitioners, consultants, or researchers in the space — provide rapid domain knowledge. These aren't your eventual users; they're people who understand the landscape from a different vantage point.

---

How to Plan Generative Research

Write a clear research question. Generative research still needs focus. "What do users want?" is too broad. "How do customer success managers communicate churn risk to their VP of Sales, and what frustrates them about it?" is a question you can actually answer.

Define your participant criteria. Avoid the temptation to interview everyone. A tight participant profile — job title, industry, experience level, specific behavior — yields richer insights than a broad one.

Draft an interview guide loosely. For generative research, your guide is a map, not a script. Know your core questions, but be prepared to follow wherever the conversation leads. The most valuable insights often emerge from an unexpected thread.

Choose your method based on constraints. Got two weeks and a budget? Consider ethnography. Got one week and a fast timeline? Async AI interviews at scale. The best method is the one you'll actually complete.

Plan synthesis before you start. Decide how you'll capture and organize data. Affinity mapping, thematic analysis, and opportunity mapping are common approaches. Koji automates much of this — extracting themes, flagging quotes, and generating insight summaries automatically after every interview.

---

How to Conduct Generative Interviews

The craft of generative interviewing comes down to a few principles:

Start broad, then narrow. Open with context-building questions: "Tell me about your role," "Walk me through a typical week." Only after establishing context do you funnel toward specific topics.

Ask about behavior, not preferences. "What would you prefer?" generates hypothetical answers. "What did you do last time this came up?" generates behavioral data grounded in reality.

Follow the energy. When a participant's tone shifts — frustration, excitement, resignation — slow down and probe. "It sounds like that part is particularly hard — can you say more?"

Use silence. After a participant finishes speaking, wait a beat before asking the next question. This pause often surfaces the most honest, unguarded responses.

Never lead. Avoid "Do you find X frustrating?" Ask "How does X feel to you?" The first plants an answer; the second invites discovery.

Koji's AI interviewer applies these principles automatically across every conversation — asking disciplined questions while adapting naturally to each participant's unique responses.

---

Synthesizing Generative Research

Raw data — transcripts, notes, recordings — isn't insight. Synthesis is where the value is created.

Affinity mapping: Group observations by theme. Look for patterns that appear across multiple participants — these become your core insights.

Thematic analysis: Go deeper than surface patterns. What underlying belief or mental model explains why multiple people do the same thing? Thematic analysis reveals the "why" beneath the "what."

Opportunity mapping: Frame insights as "How might we...?" questions. Transform "Users feel overwhelmed by notification settings" into "How might we help users maintain focus without managing dozens of preferences?"

Koji automatically surfaces themes and patterns from your interviews, generating an insights dashboard that shows which topics appeared most frequently, which elicited the strongest emotional responses, and which represent the biggest opportunity gaps. What traditionally took researchers two weeks of manual analysis takes Koji minutes.

---

Common Mistakes in Generative Research

Confirming what you already believe. It's easy to design interviews that lead participants toward your existing hypotheses. Build in deliberate checks: invite a skeptic to review your guide, alternate between directive probes and open exploration.

Talking to too few people. Five participants is appropriate for usability testing; generative research often requires 15–30 to reach thematic saturation across a segment. With AI-automated interviews, there's no reason to stop at five.

Skipping synthesis. Teams sometimes conduct interviews and move straight to solutioning based on gut feel. Proper synthesis forces rigor and reveals insights that don't surface from memory alone.

Interviewing the wrong people. Your most articulate users are not necessarily your most representative users. Make sure your participant criteria reflect the full range of experience in your target population.

Not sharing findings broadly. Generative research locked in a researcher's head doesn't change decisions. Build sharing rituals: readouts, research repositories, regular insight syncs with product and engineering.

---

Generative Research at Scale with Koji

Traditional generative research has a ceiling: one researcher can conduct 3–5 interviews per week, which means reaching 30 participants takes 6–10 weeks. This pace makes generative research impractical for fast-moving teams.

Koji removes this constraint. Your AI interviewer runs simultaneously across all participants — day or night, in any language, in text or voice mode. A study that would take a human researcher six weeks can be completed in 48 hours.

More importantly, Koji synthesizes faster too. Automatic themes, AI-generated insights, and cross-participant analysis mean you move from "I just talked to 40 people" to "here's what we learned and what we should do about it" in the same afternoon.

---

Related Resources

• Generative vs. Evaluative Research: When to Use Each Method
• Structured Questions in AI Interviews
• How to Analyze Qualitative Data: From Raw Interviews to Actionable Insights
• Customer Discovery Interviews: The Complete Guide
• In-Depth Interviews: The Complete Methodology Guide
• AI-Moderated Interviews: How Automated Research Works