The one line that makes Copilot's answers actually useful

Most disappointing Copilot answers aren't a model problem. They're a missing sentence at the start of the prompt: what it's actually looking at, and what "good" looks like.

Ask "write a summary of this project" and it'll write a plausible generic summary. Ask "write a summary of this project for our Monday steering meeting - our exec sponsor cares about budget and timeline, not technical detail, keep it under 150 words" and it writes something you could actually send.

The line to add before the actual question, every time:

• Who's reading it, and what do they already know.
• What format it needs to land in - an email, a table, three bullets, not a Copilot decides.
• Which document or dataset is the current one, if there's more than one floating around with a similar name. Copilot will happily ground itself in the wrong version if you don't say which is live.

A quick test: if you'd need to rewrite most of what came back before sending it, the prompt was missing this line, not the model falling short.

Try the same request twice - once bare, once with the audience and format stated up front - and compare what comes back. The gap is usually bigger than people expect for one added sentence.