AI UX Design: How to Design AI Features People Can Trust When the Model Is Wrong

In this guide I go through what makes AI features different from every other part of a product, the seven parts every AI feature has, and the patterns that keep a wrong answer from turning into a real mistake.

What Is AI UX Design?

AI UX design is the design of the interface around an AI feature: what people type, what they see while they wait, how the answer is shown, where it came from, how to correct it and what happens when it's useless. The model decides what the answer says. The interface decides what a wrong answer costs.

Why AI Features Need Their Own UX

Every other feature in a product does what it's told. An AI feature doesn't, and the free course AI-Product UX opens with three differences that change the design work.

• The output is probabilistic. It's usually right, sometimes wrong, and looks exactly the same either way. Confidence is a property of the writing, not of the answer.
• The input is open. A form asks for a date and gets a date. A prompt box accepts anything, so an empty box tells people nothing about what to ask.
• It's slow, and unevenly so. The same request can take 5 seconds once and 30 the next time, which breaks the assumptions behind a normal spinner.

Think of a car's navigation app. It's right almost every time, and what you remember is the one time it sent you into a closed street. Good navigation shows the route before you drive it, lets you choose another one, and recalculates without a fuss. AI features need the same three habits.

The Seven Parts of an AI Feature

The course breaks every AI feature into seven parts. Most products design the output and leave the other six to chance.

1. Input: where people ask, with examples of what works.
2. Context: what the model knows about this person and this task, and whether people can see or change it.
3. Wait: what the screen shows while the answer arrives.
4. Output: the answer itself, shaped for the task.
5. Provenance: where the answer came from.
6. Controls: edit, retry, report, undo.
7. Fallback: what happens when the AI can't help.

A quick review for any AI screen: go through the seven and mark each one as designed, default or missing. The missing ones are your backlog.

How Much Should an AI Feature Do on Its Own?

The course calls this the autonomy ladder, with four rungs: the AI suggests, the AI drafts for review, the AI acts with an easy undo, and the AI acts on its own. Each rung up saves people time and raises the cost of a wrong answer. A useful rule: move a feature up one rung only after people trust the rung below, and always keep a way back.

How Do You Explain an AI Answer?

People trust an answer they can check. The course describes four levels of explanation, from a short scope line ("Based on your last 30 days of orders") up to the highlighted passage in the source document. Pick the level by the stakes: a playlist suggestion needs a scope line, an answer about money, health or a contract needs the passage.

What Should People See While They Wait?

A spinner that runs for 30 seconds looks like a frozen product. Streaming the answer turns waiting into reading time, because people start reading almost at once. When streaming isn't possible, show progress that tells the truth and let people keep working on something else.

Designing for Wrong Answers and Refusals

The course separates four kinds of useless answer, and each needs its own words and its own action.

In the course practice, the model is wrong roughly one time in ten. Design for that one time first, and the other nine take care of themselves.

Where to Start with AI UX Design

Do you work on a feature that writes, summarizes or recommends? Would a wrong answer cost a user more than a moment? Could you list today what your screen shows when the model can't help?

If that's three yeses, start with the free course AI-Product UX: seven short lessons and a 45-minute practice, about an hour in total. Then check where you stand on the wider skills with the UI/UX skills assessment, or see how AI is changing the work in UI/UX design trends 2026.