← All guides

AI + UX · 7 min read

AI chat interfaces people trust

Design AI chats that feel honest: clear roles, easy edits, and no fake “human” vibes.

Explained like for kids

Talking to AI is like asking a very fast helper who sometimes guesses. Good chat design tells you “this is the helper,” lets you fix your question, and never pretends the helper is your best friend when it isn’t.

Examples

Each example has a kid-friendly “why,” then a clear Don't and Do with a real UI mock so you can see the difference.

Say it’s AI out loud

Like for kids: If a robot wears a human mask, you feel tricked when it messes up. Better to know it’s a robot helper from the start.

Don’t

Pretending the bot is a human support agent

Broken trust when answers are wrong. People expect humans to know your account history.

Wrong — fake human agent
S

Sarah · Support

Hi! I’ve reviewed your account and fixed billing.

Looks human — but it’s a bot guessing

Do

Clear AI label + what it can help with

Honest expectations. People know to double-check important stuff.

Right — clearly labeled AI
AI assistant · Can make mistakes

I can explain plans and draft replies. I can’t change billing myself.

Easy to fix a bad prompt

Like for kids: If you ask the wrong question in class, you should be able to rephrase — not start the whole conversation from zero.

Don’t

No edit — only “start new chat”

High interaction cost. People abandon after one bad answer.

Wrong — start over only

Write a friendlier error for empty search…

Oopsie! Nothing here lol 😅

New chat

Do

Edit message + Regenerate answer

Small fixes beat rewriting everything. Faster learning loop.

Right — edit + regenerate

Write a friendlier error for empty search…

No results yet. Try fewer filters.

EditRegenerate

Show the source of the answer

Like for kids: When a friend says “I read it in a book,” you trust them more than “I just know.” AI should show its homework too.

Don’t

Confident answer with zero sources

Looks smart, may be inventing facts. Hard to verify.

Wrong — no sources

Refund policy

You can get a full refund within 60 days, no questions asked.

Sounds sure — might be invented

Do

Cite docs / pages the answer used

People can check. Builds calibrated trust, not blind trust.

Right — citations shown

Refund policy

Full refund within 14 days if unused.

Terms §4Help → Refunds

One clear next step

Like for kids: After the teacher explains, they say “now try problem 3” — not ten different homework piles at once.

Don’t

A wall of equally loud suggestion chips

Choice overload after the answer. People freeze.

Wrong — chip overload
RewriteShorterLongerFormalFunnyTranslateEmailTweet

Do

One primary action + quieter alternatives

Hierarchy guides what to do next without noise.

Right — one primary next step
Insert into draftMake shorterMore…

Remember these

  • Label the AI clearly — never fake a human agent.
  • Let people edit the last message and regenerate.
  • Show what the AI used (files, context) when it matters.
  • Keep one primary next action after an answer.

Practical AI product patterns for designers building chat and copilots. Rewritten in plain language for learning — not a reprint of the originals.