Johnny Chan logo
AI DesignMay 18, 20265 min read

By Johnny Chan · UI/UX Designer, Hong Kong

UX Writing for AI Assistants: Starters, Replies, and Limits

How to write suggested prompts, in-thread replies, and refusal copy so assistants feel useful without sounding robotic or overconfident.

UX Writing for AI Assistants: Starters, Replies, and Limits

An assistant can look polished in Figma and still feel broken in production because the words are vague. I treat AI copy like any high-stakes UI: short, specific, and tied to what the system can actually do today.

Starter prompts should mirror real jobs

Swap “Ask me anything” for three tasks users already try in support or search: track a shipment, compare plans, summarize a policy page. Good starters teach scope faster than a feature bullet list.

Keep tone calm and specific

  • Skip fake cheer and endless apologies. Sound like a capable colleague.
  • Use plain words. Save jargon for expert modes or B2B admin tools.
  • Match the product: warm for consumer apps, crisp for finance and ops.

Say no with a path forward

When the model cannot act, explain why and where to go: “I cannot change your password here. Open Settings → Security.” That beats a generic “Something went wrong” that makes users think the product is broken.

Run five usability sessions on copy before launch. Bad phrases ship to everyone on day one.

Mine logs for repeat questions

If people ask the same thing three different ways, your first reply or starter missed intent. Designers and writers should review those threads weekly in the first month after ship.

Let's work together

Open to UI/UX projects, collaborations, and product design support in Hong Kong and remotely.

Let's Connect