Prompt & Done · BULLSEYE
The 4-part prompt: same AI, better answer
We ran the same everyday task twice through Claude on 7 Nov 2026. The one-liner returned a 220-word template with 9 blanks to fill. The 4-part prompt returned a 70-word text, ready to send, ending in one clear ask.
The four parts follow Anthropic's prompting guide (be clear and specific, give context, say the format). Works the same way in ChatGPT or Gemini. One run each; outputs vary.
1 Copy the template
CONTEXT: I'm [who you are]. [What's going on, the facts, who it's for.]
TASK: Write [exactly what you want made].
FORMAT: [Shape: a text / email / table / bullet list], [length: under N words], [tone].
LIMIT: Keep it [what to keep it to]. Instead of [what you don't want], [do this].Tip from the guide: phrase limits as what to do ("keep it friendly and plain") rather than only what not to do.
2 The exact prompt from the reel
You're helping a tenant write to their landlord. I rent a flat; the heating has been broken since Monday, it's about 12°C inside, and I emailed once on Tuesday with no reply. My landlord is Mr. Patel and we're on good terms.
Write a text message asking him to get the heating fixed.
Keep it under 80 words, friendly but firm, ending with one clear ask: a repair date by Friday.
Don't threaten or mention legal action, and don't use phrases like "I hope this finds you well".What came back: "Hi Mr. Patel, it's [Your Name] from the flat. The heating has been broken since Monday and it's down to about 12°C inside, so it's getting hard to keep warm. I emailed on Tuesday but haven't heard back, so I'm texting in case it got missed. I know repairs take some arranging, but I do need this sorted soon. Could you let me know a repair date by Friday?"
The one-liner ("write a message asking my landlord to fix the heating") gave a generic email with [Landlord's name], [address / unit number], [date it started] and six more blanks.
3 Why it works
Context tells the model the facts it can't guess. A specific task stops it hedging with a template. Format sets the shape and length. A limit removes the filler. Anthropic's test: show your prompt to a colleague with no context; if they'd be confused, the model will be too.