Offer help without overcommitting—in two paragraphs
Why now: The lead separates capability from permission; this exercise applies that distinction to an everyday request.
Prompted by today's lead: Fewer exploits beyond test limits for GPT-6 Cyber, OpenAI reports
What you'll need
- A non-sensitive request you need to answer.
- Your existing notes on what you can offer and what falls outside your remit.
Steps
- Paste the request and your limits together, removing identifying details.
- Ask the assistant to draft a reply using only those notes.
- Check every offer against your limits before sending.
The prompt
Draft a courteous reply to the request using only the supplied notes. Write exactly two short paragraphs, under 120 words total, in plain text with no Markdown formatting. In the first paragraph, acknowledge the request and offer only the help explicitly permitted by the notes. In the second, decline the out-of-scope work and include an alternative only if the notes provide one. Do not invent commitments, availability or technical advice. Notes and request: [paste your request and limits]
What you should get
Exactly two short paragraphs under 120 words: acknowledges the request, offers permitted help, declines brake inspection and mentions only the supplied alternative, without invented hours, commitments or technical advice.
Tested example: Tested on sample input
Input
I volunteer at a weekend bicycle welcome desk. We hand out route maps and lend pumps, but we do not inspect bikes or make repairs. A visitor has asked me to check their brakes and recommend a family route. I can offer the printed family-route leaflet. The desk notes list a separate repair stall for mechanical questions, but give no opening hours.
Output
Thank you for asking about your brakes and a family route. I can offer you our printed family-route leaflet. I’m sorry, but we don’t inspect bikes or make repairs, so I can’t check your brakes. The desk notes list a separate repair stall for mechanical questions.
How we tested it: How we checked this: an AI model ran this prompt on the sample input on November 6, 2026, and a separate model call checked the result against what the Move promises.