By Miguel Clavel, Senior Product Designer

AI UX patterns

Seven interface patterns for AI products people can trust. Each one has a working demo, when to use it, and what goes wrong without it. The demos are scripted, so there's no model behind them: the point is the interface, not the AI.

Pattern 1 · Transparency

Show your sources

Does my policy cover a rental car after an accident?
Yes, if you added rental reimbursement: up to $30 a day for 30 days while your car is in the shop . It doesn't apply if the car is only being inspected .
Your policy, page 4, Rental reimbursement: "We will pay up to $30 per day, for no more than 30 days, for a rental vehicle while your covered auto is being repaired."
Your policy, page 4, Exclusions: "Coverage does not apply when the covered auto is withdrawn from use only for inspection or estimate."
Use it whenThe answer is about money, health, rules or anything someone will act on.
Without itA confident wrong answer looks exactly like a right one.
Where I've used itEvery answer on my chat portfolio says it was written from my resume and case studies.

Pattern 2 · Honesty

Say how sure you are

Use it whenSome answers are facts on file and some are guesses.
Without itEvery answer sounds equally sure, so people trust the wrong ones.
The ruleLow confidence should change what the interface offers next, not just the label.

Pattern 3 · User control

Let people edit the AI's work

Suggested reply to a customer

Use it whenThe AI drafts something a person sends or signs.
Without itPeople either accept bad drafts or abandon the feature.
The ruleThe suggestion is a starting point. The person's edit always wins, and is never overwritten.

Pattern 4 · Correction and recovery

Go back a step without starting over

    Use it whenA conversation or flow collects answers one at a time.
    Without itOne wrong tap means starting the whole thing again.
    Where I've used itThe conversational insurance quote I designed keeps every answer editable at every stage.

    Pattern 5 · Trust

    Say why you're asking

    What's your date of birth?
    Why we askInsurers price by age band. We use it only to compare quotes, and you can see exactly who receives it before you continue.
    Can we pull your driving record?
    Why we askIt makes the quote match what you'll actually pay. It's optional: say no and you'll get an estimate instead.
    Use it whenThe question is personal, regulated or surprising.
    Without itPeople drop off at exactly the question that matters most.
    Where I've used itThe conversational quote asks for consent at the point of collection, not in a wall of text at the start.

    Pattern 6 · Clear boundaries

    Say what it can't do, and hand off well

    Can you cancel my policy and refund me?
    I can't cancel policies or move money. A licensed agent can do both, usually within a day. I'll pass along what we've covered so you don't have to repeat it.

    Use it whenThere are things the system must never do on its own.
    Without itIt tries anyway, or refuses with no way forward.
    The ruleA boundary always comes with a next step, and the context travels with the handoff.

    Pattern 7 · Human oversight

    Show exactly what will happen, before it happens

    I've drafted your request. Here's exactly what will be sent:
    To
    Claims team
    Message
    Requesting a rental car under rental reimbursement while my car is repaired. Claim opened today.
    Attached
    Repair estimate (1 file)

    Use it whenAn agent is about to act on someone's behalf: send, pay, book, delete.
    Without itPeople find out what the AI did after it's too late to stop it.
    Where I've used itThe paint tool on my second portfolio shows your message, name and email before anything is sent.