The PROMPT Framework: Six Levers for Better AI Answers

Most people type a few words into ChatGPT or Claude and hope for the best. That habit comes from Google, where a handful of keywords is all you need. AI works differently. It predicts the most likely response to your input rather than searching for an existing answer. Vague input leads to vague output.

Writing a good prompt is closer to briefing a new hire than typing a search query. The clearer the brief, the better the work. The PROMPT framework offers six levers to pull when the output misses the mark. You rarely need all six, but knowing they exist is the difference between a tool that saves an hour and one that wastes ten minutes.

There are two P's: Persona and Provide Context. This is intentional. They perform distinct jobs and both deserve their place.

PROMPT at a glance

Letter	Means	What it does
P	Persona	Sets the AI's role
R	Reasoning	Makes it think first
O	Output Format	Sets structure, length, tone
M	Multiple Examples	Shows the standard
P	Provide Context	Supplies the facts
T	Task Clarity	The clear ask the rest produces

I will use one running example. You own a small hair salon and want to email clients who have not booked in a few months to bring them back. Watch how each lever sharpens the request.

P — Persona

Assign the AI a role before you ask for anything. A role sets the vocabulary, assumptions, and depth. A copywriter who specializes in local service businesses produces sharper work than no role at all because it pulls the model toward the right patterns immediately. This is the fastest single upgrade you can make to a prompt.

Before: Write an email to get old salon clients to come back.

After: You are a copywriter for small local service businesses. You know how to win back lapsed customers without sounding desperate or spammy.

R — Reasoning

Left alone, AI jumps straight to an answer. Ask it to think first: the answer improves, and you can see its logic. If something is off, you spot it immediately instead of after you hit send. For a win-back email, that means considering why clients drift away before writing a single line.

Before: Write the email.

After: Before writing, think through why a salon client stops booking, what would make them return, and what tone avoids sounding pushy. Then write the email.

O — Output Format

AI defaults to a wall of text. Decide the shape of the answer before it begins; reshaping it later usually makes it worse. Set length, tone, and structure in one sentence. This is the most underused lever, and it costs nothing.

Before: Write a win-back email.

After: Write a short email, under 120 words. Friendly, not salesy. Include a subject line and the body. End with one clear call to action to book online.

M — Multiple Examples

Descriptions leave room for guessing. An example ends the guessing. Paste a previous email that landed well, and the AI matches your voice rather than inventing a generic one. One real example beats a paragraph of instructions about tone.

Before: Make it sound like us.

After: Here is an email we sent last spring that got a lot of replies: [paste it]. Match that tone and rhythm.

P — Provide Context

This lever fixes most weak answers. When details are missing, AI fills the gap with a confident guess. Every fact you add removes one guess. If the result feels generic, the fix is more context, not a cleverer question.

Before: Write an email to past clients.

After: We are a three-chair salon in Philadelphia. These clients last visited three to six months ago. We just added a new colorist with open slots. Our average cut is $65, and we are not running a discount.

T — Task Clarity

Task Clarity is not a sixth trick you bolt on. It is what the first five letters produce. Persona told the AI who it is. Reasoning set how it thinks. Output Format fixed the shape. Multiple Examples showed the standard. Provide Context supplied the facts. Once those are in place, the task is obvious, and you can state it in one clean line.

It works on you, too. Filling in the first five forces you to decide what you actually want, which is usually the hard part. By the time you reach T, the ask has written itself.

So the test is simple. If the instruction still feels hard to write, the instruction is not the problem. One of the earlier five is missing, and you are asking the task line to do work that belongs upstream.

Before (nothing set up): Help me with a salon email.

After (P, R, O, M, P already in place): Write the email. Do not invent discounts I did not mention, and ask first if you need anything from me.

The instruction got shorter, not longer. The setup carried the weight, so the ask did not have to.

The loop: treat the first draft as a rough draft

Prompting is a cycle. Generate, read critically, refine, and run it again. The second or third version is what you actually use. Reading that first draft reveals what you forgot to say.

Try these follow-ups:

"Cut it in half and make the opening line stronger."
"The tone is too salesy. Make it warmer and more personal."
"Give me three subject line options."

Setup prompts vs. task prompts

You do not need to cram all six levers into one giant message. For anything beyond a quick question, use two stages.

Start with a setup prompt. Use PROMPT to define who the AI is, what it knows, and how it should work. You only write this once.

Then send task prompts. One focused request at a time, inside that context. Write the email. Turn it into a text message. Draft a social post. Each builds on the setup.

If you stack more than two unrelated jobs with "and," split the prompt.

A full example

Here is the framework in one setup prompt. Once it becomes habit, drop the labels and write it as a paragraph.

[P] PERSONA
You are a copywriter who writes for small local service businesses. 
You win back lapsed customers without sounding desperate or spammy.

[R] REASONING
Before writing, think through why a salon client drifts away, what 
brings them back, and what tone avoids sounding pushy.

[O] OUTPUT FORMAT
Short email, under 120 words. Friendly, not salesy. Give a subject 
line and body. End with one clear call to action to book online.

[M] MULTIPLE EXAMPLES
Match the tone of this email that worked for us: [paste your best 
past email].

[P] PROVIDE CONTEXT
Three-chair salon in Philadelphia. These clients last visited three 
to six months ago. New colorist with open slots. Average cut $65. 
No discount offered.

[T] TASK CLARITY
Write the email. Do not invent discounts. Ask first if you need 
anything from me.

Where to start

You will not reach for all six levers every time. A quick question might only need Persona and Task Clarity. The goal is to know which lever to pull when an answer misses the mark.

If you only remember two, use Provide Context and Output Format. They fix the most for the least effort. Add the rest as you go, and let the loop do the polishing. Aim for a fast second prompt rather than a perfect first one. Do the first five well, and Task Clarity comes nearly for free.