40 components in 8 categories
AI UX patterns: trust, steering and feedback
Much of what makes an AI product feel trustworthy is small: showing why it suggested something, letting a person steer while it works, and making a wrong answer easy to correct. These patterns come from the AI Essentials set and cover trust and privacy (mask personal data, temporary chat, note taker consent), steering (how hard to think, answer as a list, table or text) and feedback (correct a number, which answer is better, learn from my edits).
They also cover what happens around an answer (copy as, share this chat, read it aloud, send to an app), how a conversation starts (questions before it starts, pick up where you left off), making the assistant your own, search by meaning and working together (catch me up, action items from a meeting). Each is a single interaction, small enough to add to a product in a day.
Each one is three small files (HTML, CSS and vanilla JS, no framework) and a prompt that rebuilds it in Claude, Cursor, v0 or Lovable. All of them are free with an account.