Write guardrail rules for a returns chatbot

About this exercise

You're writing the rule list for your store's returns chatbot. Every line must start with "Always" or "Never", at most 8 bullets, plain text. The return window changed to 30 days on Oct 1 even though the policy doc still says 60. The bot must hand off to a human when a customer mentions a chargeback, a lawyer or an injury, and must never promise a refund over $50 itself.

Common mistakes when you ask AI for this

• Writing a long system prompt. The AI writes paragraphs of persona and tone advice instead of rules a bot can follow. Try: Say: "Every line starts with Always or Never."
• Copying the old window. The doc still says 60 days, and the AI has no way to know it changed. Try: Tell it: "The return window is now 30 days, not 60."
• No escalation rule. The AI doesn't know which topics are too risky for a bot. Try: Add: "Hand off to a human for chargebacks, lawyers or injuries. Never promise a refund over $50."
• Letting the list sprawl. AI pads lists with extra bullets to look thorough. Try: Say "at most 8 bullets".
• Getting markdown back. AI adds bold and headings by default, which show up as stray symbols where this is pasted. Try: Say "plain text, no bold, no headings, no asterisks".

What a strong prompt includes

• Sets the line format
• Gives the new window
• Names the handoff triggers
• Caps refunds
• Limits the bullets

Practice this now ↑

Related exercises

• Accept the buyer's counteroffer on three stated conditions (Sales AI · Sales and customer success)
• Accrue goods received but not invoiced at PO price (Finance AI · Accountants)
• Add dark mode by overriding your existing CSS variables (Coding AI · Software developers)
• Add input checks to your number guessing game (Coding AI · High school students)
• Add labels to a signup form without rewriting the page (Coding AI · High school students)