Conducting User Interviews for AI Features
What you'll learn: How to structure user interviews that uncover the real workflows, pain points, and trust dynamics that determine AI feature success.
Why This Matters
Understanding AI user expectations (lesson 1869) gives you the "what"; user interviews reveal the "why" and "how." Unlike traditional features, AI introduces uncertainty, opacity, and shifting mental models—making interview technique crucial to product iteration.
Core Interview Structure
1. Workflow Mapping
Start with "walk me through how you currently do X"—not "what would you want in an AI tool?" Watch for workarounds, tool-switching, and points of friction. These reveal where AI can genuinely add value versus where it's a solution looking for a problem.
Analogy: You're a detective reconstructing a crime scene, not a salesperson pitching a product.
2. Pain Point Prioritization
Ask: "What takes the most time?" and "What's most frustrating?" Distinguish between annoying (formatting) and critical (decision-making). AI solves different classes of problems differently—generation vs. analysis vs. automation.
3. Trust Factor Exploration
This is unique to AI features. Ask:
- "How would you know if the output was wrong?"
- "What would you need to see to trust this?"
- "When would you override the AI?"
Users often can't articulate trust needs upfront, but scenarios reveal them: "If the AI suggested firing this employee..." surfaces validation requirements immediately.
4. Post-Use Follow-Ups
After users try your AI feature (even prototypes), ask: "What surprised you?" and "When did you stop trusting it?" Mismatches between expected and actual behavior are gold.
Anti-Patterns to Avoid
- Leading questions ("Wouldn't it be great if AI could...")
- Hypothetical scenarios without current-state context
- Skipping non-users (they often have the clearest unmet needs)
Key Takeaway: Effective AI user interviews focus on actual workflows, concrete pain points, and trust boundary conditions—not feature wish lists. Your goal is understanding context that metrics alone can't reveal.