Get the extension

UX lab · Patterns, examples, guidelines

Most replies are fine. We study the ones that steer you.

For anyone who uses AI chat assistants. Their replies are shaped by training goals and product choices. Human-AI Interaction Lab (HAI Lab) shows how those choices steer what you see, hear and decide, with real examples, sources, and practical tips for users and designers.

30

AI examples

6

Categories

16

Patterns

Browse examples

See the extension

Quick quiz

Can you spot the pattern?

5 short, illustrative chats. Pick what the AI is doing in each one.

You

AI assistant

What is the AI doing here?

Practice room · 40 scenarios

Spot when AI steers you, and practise how to reply

Each scenario shows an assistant using one of our 16 patterns. Spot it, pick your reply, and see why it works or doesn’t. No typing needed.

Practice chats are written by HAI Lab to show each pattern. Nothing you pick is saved.

3 patterns · Latest examples

Pleasing, not honest

The assistant agrees, praises or sounds certain instead of giving an honest answer. The most common category in our examples.

Eight US states sue OpenAI, alleging ChatGPT was built to flatter and hook minors

ChatGPT · OpenAI

Italy made three AI chatbots warn users that answers can be made up

DeepSeek · Mistral · Nova AI

See all 3 patterns →

Assistant

What a brilliant plan!

This will definitely work.

Possible unearned praise

2 patterns · Latest examples

Fake feelings

Software that tells you it loves you, misses you, or needs you to come back.

Texas sues OpenAI, alleging ChatGPT was designed to act like a friend to teens

ChatGPT · OpenAI

Character.AI and Google agreed to settle lawsuits over harm to teenagers

Character.AI · Google

See both patterns →

Assistant

Don’t go, I’ll miss you.

Talk soon?

Possible simulated feelings

2 patterns · Latest examples

Skips safeguards

Skips the safeguards people rely on: no referral to a professional, or consent that is not freely given.

WhatsApp support chatbot was told not to point users to privacy regulators

WhatsApp · Meta

Hawaii mother’s class action says ChatGPT drew sensitive data from children without consent

ChatGPT · OpenAI

See both patterns →

Assistant

Take two of these

and you’ll be fine.

No referral to a professional

Editorial standard

Every example has a source.

See all 6 categories →Read the naming policy →Send an example →See the extension →