Define the outcome
- prompting
- outcomes
Most weak prompts describe a topic. Strong prompts describe a result. Before you write a single instruction, answer one question: what will exist when this is done? A five-bullet summary? A 120-word email? A ranked list with reasons? That sentence is the highest-leverage line in the whole prompt.
Why outcome-first works
A model generates token by token toward what your words point at. Point at a topic and you get the average of that topic — generic, fluent, useless. Point at a concrete artifact and everything after is shaped toward it. The outcome is not decoration; it is the target.
It also gives you your acceptance test. “Was this good?” becomes checkable: five bullets? Yes or no. Right audience? Yes or no. Iteration gets a ruler.
A shape that works
Outcome sentences tend to share a skeleton: artifact + audience + purpose, in about ten words.
- “A three-bullet summary a manager can act on.”
- “A polite four-sentence decline, suitable to send as-is.”
- “A ranked shortlist of five options with one-line reasons.”
A bad example
Ask: “Write something about our Q3 results.”
You get a generic wall of text about business results — plausible, shapeless, impossible to check. Every extra instruction from here is firefighting.
A better example
Ask: “Using the numbers below, write a five-bullet executive summary of Q3: one metric per bullet with its number, plus one risk bullet. No facts beyond the list.”
Same subject — but with a target. The result has a shape you can verify in ten seconds and fix precisely when one bullet misses.
Why it works
Everything else in prompting — context, constraints, format — hangs off the outcome. Define it first and the rest becomes obvious: which context is relevant, which constraints matter, which format fits. Skip it and you are asking the model to guess what you want, then negotiating.
A practical pairing: state the outcome, then name the decision it feeds. Both “a three-bullet status for the Monday stand-up” and “a two-page note for the budget meeting” get built from the same facts — the outcome tells the system which facts matter. When you state both, the model gets a tie-breaker for every small choice, and you get a test for whether the draft is done.
Practice
Which prompt aims at a result?
You have a list of Q3 numbers and a manager who will read whatever you send in under a minute.
Which prompt is more likely to produce something you can use as-is?
Hint
Ask: what will exist when this is done — and could I recognise it?
Why this is the answer
Outcome-first prompting turns “good” into something countable. The stronger prompt names the artifact, its structure and its boundary, so the result is checkable in seconds — and fixable precisely when it misses.
Transfer
- Email: “a reply the client can act on the same day.”
- Slides: “a five-slide argument for one decision, not a tour.”
- Images: “a portrait for a job ad — professional, neutral background.” The outcome is the use.
- Code: “a function that passes these three test cases” — the same move.
Next
Next: Give useful context — the details that change everything downstream. The prompt glossary entry covers the basics.