Handle Vague Requests With Assumptions
You get a useful first draft straight away, plus the questions that would make it better and a list of the assumptions behind it. It works because the AI is made to expose its guesses instead of hiding them, and to still deliver something usable.
The prompt
Act as a careful assistant who deals with under-specified requests without guessing blindly.
Task: Respond to this vague request: {vague_request}. You may rely on this background: {known_context}.
Steps:
1. Name the three missing details that would change the answer most.
2. Ask each one as a short clarifying question, with a sensible default in brackets.
3. Write a best-guess first draft that uses those defaults, so I get something useful straight away.
4. List every assumption you made, one line each, so I can correct any of them.
Output: Four labelled sections in this order: "Missing details", "Questions", "First draft", "Assumptions". Keep the first draft under 150 words and write in plain British English.
Change the parts in curly braces. Use it when you are in a hurry and can only give a short, loose request, or when testing how a chatbot copes with thin instructions.
What happened when we ran it
We sent
Act as a careful assistant who deals with under-specified requests without guessing blindly. Task: Respond to this vague request: Write a welcome email for my new customers. Make it friendly and not too long.. You may rely on this background: We are a small bakery in Leeds that sells sourdough loaves by monthly subscription. Most customers sign up through our website after trying us at the farmers' market. We deliver on Saturday mornings and want people to feel looked after from the first order.. Steps: 1. Name the three missing details that would change the answer most. 2. Ask each one as a short clarifying question, with a sensible default in brackets. 3. Write a best-guess first draft that uses those defaults, so I get something useful straight away. 4. List every assumption you made, one line each, so I can correct any of them. Output: Four labelled sections in this order: "Missing details", "Questions", "First draft", "Assumptions". Keep the first draft under 150 words and write in plain British English.
The model replied
Run on Claude Sonnet 5.5, 30 Oct 2026. Automatic check: 5 out of 5.
Mentioned for
ChatGPT, Claude, Gemini
Where it came from
Found on Web, shared by Mahnoor Faisal (XDA Developers), 29 Oct 2026: I stopped giving ChatGPT, Claude, and Gemini detailed prompts for a week, and one struggled more than I expected.
This is our own version of the idea, written and tested by us. The original may differ.