"What does done look like?" belongs in the prompt.

A "what does done look like" prompt line is the finish line written in plain English. You can give the model a clean job, real context, a solid why, and tight rails, and still get an answer you cannot ship. The work sounds finished. It is not. You never said what finished meant.

Prompt Fluency is communicating so the model does not have to guess. "Done" is one of the biggest blanks. Humans carry a picture of done in their head. Short enough for a phone. Safe for a nervous parent. Ready to paste into the newsletter. One page. No jargon. The model only has the words. If those words never show up, it invents a finish line that looks professional and still misses yours.

Try this ask:

"Write a signup note for Saturday youth basketball."

You get a polite paragraph. Ages are fuzzy. Time is buried. The rain plan is missing. The model did the job you typed. It did not do the job you meant, because done stayed private.

Same break shows up everywhere.

"Draft the PTA fundraiser email."

"Write the bakery weekend specials post."

"Rewrite the apartment newsletter blurb about the garage sale."

Each one sounds complete. None of them say how you will judge the answer. So the model optimizes for "sounds helpful." You needed "ready to send."

Put the finish line in the prompt. One short block is enough.

"Write the youth basketball signup note. Done looks like: ages 9 to 11 on the first line, Saturday 10am at the north gym, rain moves indoors, reply-by Friday, under 100 words, calm tone for first-time parents."

"Draft the PTA fundraiser email. Done looks like: goal amount in the first two sentences, volunteer shift link once, no guilt language, under 150 words, something a busy parent can skim on a phone."

"Write the bakery weekend specials post. Done looks like: three items max, prices included, Saturday hours only, no hype adjectives, ready to paste into Instagram without rewriting."

"Rewrite the garage sale newsletter blurb. Done looks like: date, start time, where to drop donations, what not to bring, under 80 words, no cute jokes."

That last line is not extra polish. It is the measuring stick. Context tells the model the room. Intention tells it the why. Constraints put up the rails. Missing details fill the blanks. "Done looks like" tells it when to stop.

I am not asking you to write a contract. I am asking you to steal one question from real work. Before you send a teammate a task, you already know what acceptance looks like. Use that same sentence with the model.

A quick check before you hit send:

Job vs done

• Job: what to produce
• Done: how you will know it is shippable
• If you can only say the job, you have not finished the prompt
• If two good answers could both "count," your done line is still fuzzy

When the answer comes back full and still useless, do not stack soft follow-ups. Ask yourself what done looked like in your head. Type that. Send again.

You're not being picky. You're removing the need to guess. That's Applied Intelligence Mastery in practice: the finish line in the prompt, not only in your head.