Library › Research AI › Customer support teams
Catch errors in a support bot's warranty answer
Practice: Catching mistakes and made-up facts. About 5 minutes.
About this exercise
Your help center bot answered a customer's warranty question, and you're reviewing it before it's used as a training example. You want only the errors, as numbered lines in the form "Bot says X; warranty says Y", under 60 words total, plain text. Use only the warranty terms, skip lines that match, and don't rewrite the answer.
Common mistakes when you ask AI for this
- Asking for a better answer. The AI writes a corrected reply, and you lose the list of what the bot got wrong. Try: Say: "List only the errors. Don't rewrite."
- Not naming the source. The AI may judge by typical e-bike warranties instead of yours. Try: Add: "The warranty terms are the only source of truth."
- Flagging correct lines. The AI confirms every line to look thorough, burying the errors. Try: Say: "Skip lines that match."
- Trusting the AI to count words. AI is bad at counting. Ask for a total word limit and it often lands 10-20% over. Try: Cap each line: "each line under 13 words."
- Getting markdown back. AI defaults to bold labels and headings, which show up as stray symbols in the training doc. Try: Ask for "plain text, no bold, no headings."
What a strong prompt includes
- Names the source of truth
- Asks for errors only
- Gives the line format
- Sets a length limit
- Asks for plain text
Related exercises
- Alert sales to a competitor's price change, facts only (Research AI · Product managers)
- Answer a buyer's HOA questions from the bylaws (Research AI · Real estate agents)
- Answer a coworker's travel policy question in one Slack reply (Research AI · Office and admin)
- Answer a family's visiting questions from the current policy (Research AI · Nurses and healthcare workers)
- Answer winter break housing questions from the current policy (Research AI · College students)