Generative AI is about to revolutionize user research.By tapping into the shared knowledge of humanity, we're gaining a powerful new tool to understand users.Over the past few months, I've been exploring interviewing ChatGPT as if it were a real user or stakeholder. The experience has been fascinating, revealing both tremendous opportunities and some fundamental principles for successful 'artificial interviews.'Here are some principles that I've found effective:

• Set Goals: Make your intentions clear to ChatGPT. For example, say, "I will interview you for a project about fitness." You can also give it more context, like "I'm interested in exploring the joys and challenges people have with fitness".
• Define the Persona: Give ChatGPT a specific persona to embody. For example, "You're a 42-year-old mother of two from New York, passionate about wellness and balancing a busy career with family life. You're also interested in technology. You live in Brooklyn and work as an executive in a creative agency".
• Set the Style: Request human-like responses from ChatGPT by saying, "You'll answer my questions one by one like an intelligent human, not like a machine. Add subtle human touches of humor in your responses "
• Ask Thought-Provoking Questions: Ask open-ended questions and probe deeper into the conversation, just like you would when interviewing human participants. As with any interview, the more thoughtful questions you have, the higher quality answers you'll typically get.
• Explore Multiple AI Personas: Interview several AI personas, analyze the responses, and look for interesting patterns. Include 3-5 different personas that represent the relevant user group of your offering. You can even use a separate ChatGPT thread to synthesize your findings.
• Validate Your Solutions: Use the same thread to test your solutions and receive feedback from your synthetic participants. Describe your solution to them to elicit reactions.
• Complement Human Interviews: Remember, while AI is a powerful tool, it can't replace the richness of human interviews. Use it to enhance your research, not replace it.
• Beware of Bias: Keep in mind that AI learning data can carry inherent biases, which could potentially skew the responses you receive. Also, languages with less speakers (like my native Finnish) can be more challenging to draw good answers from as the learning data set is smaller.

However, there's a fundamental risk in treating AI-generated responses as genuine user insights: these personas lack lived experience, emotional authenticity, and the unpredictable contradictions that make human behavior so complex and context-dependent. Organizations that over-rely on synthetic interviews may develop products based on statistically average responses rather than discovering the surprising edge cases, unmet needs, and genuine pain points that only emerge through real human connection.

To wrap up, artificial interviews provide a fascinating new approach to user research. They can complement human interviews, serve as a practice run before real conversations, and give us a fresh perspective on user behavior and preferences.

‍

‍