A surprising amount of whether an AI feature succeeds has nothing to do with the model and everything to do with the interface around it. Usability research — much of it from the Nielsen Norman Group's ongoing work on AI UX — points to a consistent set of patterns that earn trust, and a few that quietly destroy it.
Why AI UX is different
Traditional software is deterministic: the same input gives the same output, so users build a reliable mental model. AI is probabilistic — it can be wrong, vary between tries, and sound equally confident either way. Good AI design is largely about helping users calibrate their trust: trusting the system when it's reliable and staying skeptical when it isn't.
Patterns that build trust
- Show your sources. When an answer cites where it came from, users can verify it — and forgive the occasional miss. This is a big part of why retrieval-based systems feel trustworthy.
- Signal uncertainty. A system that can say "I'm not sure" or surface a confidence cue beats one that's always certain — especially when it's wrong.
- Make output easy to edit, not just accept. Treat AI as a draft the user refines, not a verdict they must take or leave.
- Keep the human in control. Preview before action, undo, and clear "are you sure?" moments for anything consequential. (More on human-in-the-loop design.)
Patterns that destroy trust
- Confident wrongness with no escape hatch — no way to correct, flag or undo.
- Hiding that it's AI, then being caught — one obvious error erases credibility.
- Over-automation, where the system acts before the user is ready.
Users don't expect AI to be perfect. They expect to stay in control when it isn't. That single principle resolves most AI UX decisions.
The takeaway
The model is only half the product. The interface decides whether people trust it enough to keep using it — which makes design a core part of any serious AI build, not a coat of paint at the end.
Sources
- Nielsen Norman Group — AI & Machine Learning UX research