"You are an expert" is not a prompt.

A "you are an expert" prompt feels like you did the hard part. You gave the model a title. Expert copywriter. Expert scheduler. Expert inventory planner. Then the answer still misses the mark, and it feels like the tool failed. It did not fail. You never told it the job.

Role lines are the ceiling in a lot of template courses. Assign a persona. Paste a framework. Hit send. That is not Prompt Fluency. Prompt Fluency is communicating so the model does not have to guess. A job title is not that communication. It is a costume.

Try this ask:

"You are an expert event planner. Build me a Saturday plan."

The model will sound confident. It will invent a timeline, a budget, and a vibe. None of that had to match what you meant. A neighborhood soccer camp for 7 to 9-year-olds. Rain plan. Parent drop-off at 8:30. No apps. The role never carried those facts. The model filled the blanks with something statistically reasonable.

Same break shows up at work.

"You are an expert operations manager. Fix the snack schedule."

"You are a senior analyst. Price the lawn route."

"You are a bookstore consultant. Sync the inventory."

Clean language. Empty job. Which schedule. Which route. Print vs ebook SKUs. The model will produce something that looks managed. Looking managed is not the same as matching the real constraints.

Strip the costume and put the work in the prompt instead.

"Build a Saturday plan for a neighborhood soccer camp. Ages 7 to 9. Drop-off at 8:30am. End by noon. Include a rain plan that stays outdoors under the pavilion. No app downloads for parents. One page."

"Rebuild the volunteer snack schedule for March home games. Keep the current rotation. Rivera family is out the second week. Output a simple table, not a speech."

"Price a weekly lawn-care route for one HOA. Eighteen yards. About 40 minutes per stop. Give a plain monthly cost and the assumptions you used."

"Sync Harbor Books inventory so print and ebook SKUs stay separate. Flag mismatches. CSV only."

No expert hat required. The model now has a job, a why, rails, and the boring facts. That is the same stack as context, intention, constraints, and missing details. The role line did none of that work for you.

I am not saying roles are useless. A short role can set tone when the rest of the prompt is already clear. "Write as a calm ops lead" after you already named the schedule, the exception, and the format. Fine. Lead with the role and hope the rest is implied, and you are back to guessing.

A quick check before you hit send:

Role vs job

• Role: who the model pretends to be
• Job: what you are actually trying to accomplish
• If the prompt only has the role, you have not prompted yet

When an answer comes back polished and wrong, do not keep chatting hoping the "expert" wakes up. Rewrite the ask. Name the outcome. Say the why. Put the rails in. Type the obvious details. Then send again.

You're not hiring a persona. You're removing the need to guess. That's Applied Intelligence Mastery in practice: communication first, costume second (or not at all).