Why it's worth the line
Most prompts describe the activity ("write an email") but not the result ("get a reply"). The model optimizes for whatever it can see — so it nails the activity and whiffs the outcome you never named.
A stated goal is also your scoring rubric. Without it you can't say whether the output worked; you just react to vibes and re-roll.
How to catch it in your own prompt
- The prompt names a task but no result — what should be true after someone reads/uses this?
- You'd struggle to write the one-line test for "did this work?"
- Two reasonable outputs could both "complete" the task while only one achieves your actual aim.
What to write instead
- Add a "so that…" clause: the task, then the outcome it serves.
- Name the single success metric — a reply, a signed doc, a passing test, a decision made.
- If there's a failure you're trying to avoid, say it; "good" is often defined by its opposite.
Seen side by side
Help me write a cold email to this lead.
Write a cold email whose only job is to book a 15-min call — success is a reply, not a sale.
Why it lands — "A reply, not a sale" stops the model from cramming the whole pitch in and killing the open rate.
Summarize this incident report.
Summarize this incident report so an on-call engineer can decide in 30 seconds whether to page the team — lead with severity and blast radius.
Why it lands — The goal ("decide whether to page") reorders everything: severity first, narrative last.
Stop fixing this by hand.
Paste a prompt — Meerkat catches this and the seven other weak spots before you send it. Free, no signup.