Skip to content
Building an AI-Powered Product or ServiceLesson 3 of 13

3 · Designing the experience (UX) around AI

A great AI product isn't "a model with a text box." It's an experience that helps a real person get a job done and feel in control. Google's People + AI Guidebook — a practical playbook from a team that designs AI products — is built on one idea: design for the human, set honest expectations, and earn trust (PAIR, 2021).

Founder-level UX principles for an AI feature:

  • Set expectations up front. Tell users what the feature can and can't do, in plain words. Over-promising ("AI does it all!") guarantees disappointment; honest framing ("AI drafts it, you review") builds trust. PAIR calls this onboarding users to AI's abilities and limits.
  • Make AI's role obvious. The user should always know when they're getting AI output versus a fact, and that it might be wrong. Surprise AI feels like deception.
  • Keep the human in control. Let people edit, undo, accept, or reject what the AI produces. A draft they can change beats an answer they're stuck with.
  • Show your work when it matters. If the AI makes a claim, let the user see where it came from (a source, the document it read). That's the Learn.WitUS trust pattern in product form — and PAIR's "explainability and trust" guidance.
  • Design the failure, not just the success. What happens when the AI is unsure or wrong? A graceful "I'm not certain — here's how to get help" beats a confident lie.

The biggest beginner mistake: shipping the raw model. A model behind a text box, with no expectations set, no editing, no sources, and no fallback, feels like magic for five minutes and like a scam by minute six. The product is everything you wrap around the model.

Check yourself. Name two things a good AI experience does for the user (besides give an answer) — and why "just a text box on a model" usually fails.

Sources

Sign in to track your progress