Researching AI products: Key insights from Mo

Over the past few months, we’ve spent a lot of time working on Mo, Alan’s AI health assistant. And it pushed us to ask ourselves: what does it mean to do research on an AI product?

The methods haven’t changed drastically; we still use interviews, observation, conversation analysis, and quantitative data.

What changed is the type of questions we need to ask and the dimensions we focus on. With AI, unpredictability, personality, and fast-changing perceptions shape the user experience in ways we don’t usually encounter in more traditional product experiences.

Research showed that members already trust Alan as their health partner. People know our content is built with doctors and that our guidance is reliable. This gives Mo legitimacy from the first interaction.

But trust also has limits.

Members worry about two distinct things:

Research helps us understand these boundaries, identify where reassurance is needed, and ensure transparency at every step.

Understanding these expectations is what allows Mo to feel credible and safe, especially in sensitive moments.

AI is polarizing, and perceptions and expectations evolve fast.

We’ve seen skeptical users pleasantly surprised by Mo, especially compared to their previous experiences with other AI tools. But we’ve also learned that people struggle to project themselves. With AI, some usage patterns don’t exist yet.

Members who had never considered asking health questions to AI tools or had never felt at ease with them began to feel comfortable with Mo, sharing skin pictures and lab results.

Research helps us track evolving perceptions, but building the right product also requires a strong product vision, because stated preferences don’t always predict future behavior.

With many companies launching AI assistants, the key question becomes: What lasting value does Mo create in members’ real lives?

To answer that, we look at what members actually do. Conversation analysis reveals surprising, sometimes unexpected interactions.

Lasting value comes from real-life usefulness, from helping members navigate health moments with clarity and confidence, not from the “wow effect” of AI.

AI isn’t just about answers; it has a personality. Some members want warmth and reassurance. Others prefer concise, efficient responses.

We saw clear differences in how members address Mo: some call it by name (in about 12% of conversations, the user uses “Mo”), while others adopt a more formal tone. These interactions reveal how Mo’s personality lands.

Fine-tuning tone ensures conversations feel engaging, respectful, and aligned with sensitive health topics.

Mo isn’t a standalone tool. It’s designed to be a central companion, guiding members seamlessly across Alan’s broader ecosystem: the medical chat, content, and support.

Every interaction counts. A wrong redirection can break the experience. Research shows how members navigate the app, where they struggle, and how smooth transitions feel.

When a member moves from Mo to a doctor or customer support agent, a smooth handoff is critical, especially if they’re anxious or in a hurry.

By becoming a trusted companion for navigation, information, and proactive support, Mo doesn’t just answer questions; it also reinforces the quality of the entire ecosystem, connecting members to the right services at the right moment. For instance, Mo drives 30% of the teleconsultation traffic.

Researching AI products requires a different lens. They help us capture not only what Mo does today, but also what it could become: an assistant that is reliable, caring, and genuinely meaningful in moments when health matters most.