Custom instructions with the 4Ps

The 4Ps are not another formula for writing prompts. They are a way to configure your AI tool's custom instructions, once, so every conversation starts with the basics already said.

If every time you open a conversation you have to repeat who you are, how you want answers and which rules to follow, you are doing every day a job that gets done once. The 4Ps organize that job into four pieces.

The four Ps

P	What it is	Question it answers	Typical mistake
Profile	Who you are and what you do	What does the AI need to know about you to calibrate its answers?	Dumping a long biography with no function. Only what changes how it should answer you goes in
Persona	The role the AI adopts	Who do you want it to be when it answers you?	A decorative role ("you are an expert"). With no experience and no point of view, it changes nothing
Protocol	The rules of behavior	What should it do when information is missing, doubts come up or a call has to be made?	Vague rules ("be helpful") or rules written as prohibitions that leave everything else open
Presentation	The output format	What does a good answer look like to you?	Not saying. Without a format, the AI picks one, and it almost always picks long

A full example

This is an illustrative example to show the structure. Replace it with your own case.

Profile: I'm a sales director at a consumer goods company. My audience is sales managers and the executive committee. I work with sales figures, margins and distribution.

Persona: Act as a senior colleague in commercial strategy: direct, with judgment, who questions my assumptions instead of agreeing with me.

Protocol: If a piece of data is missing to answer well, ask me before assuming. Separate what the data says from what you are inferring. If a figure is not in the material I gave you, say so instead of estimating it.

Presentation: Write in short paragraphs and plain English. If you compare options, use a table. One page maximum, unless I ask for more.

Look at how the Protocol and the Presentation are written: they say what to do, not only what to avoid. Anthropic frames it this way in its best practices guide: instead of "do not use markdown," ask for "write in flowing paragraphs." The negative instruction leaves everything else open, and the positive one defines the destination.

And in the Protocol, keep ALWAYS and NEVER for what is truly non-negotiable. OpenAI recommends the same, and adds that rules that contradict each other create more instability than missing details. One rule, once.

Where each P lives

Every tool names things differently, and screen names change often. What does not change is the function of each piece. I take Claude as the example, according to its current help center:

P	Where it goes	Why
Profile	The global instructions field, for the stable part. What builds up with use is saved by Memory. What is specific to a client or project goes in the project instructions	The global field applies to all your conversations. A project's instructions only apply inside that project
Persona	The global field, or a skill if you want to be able to swap it	A skill is swappable, the global field is fixed
Protocol	The global field	It is what that field describes: your preferred methods and your general communication instructions
Presentation	The global field, or a skill	The documentation says skills can adjust tone and format

Two clarifications. Memory (Settings, Memory) is on by default in the Free, Pro and Max plans, and it automatically saves your role, your professional context and your communication preferences. And each project has its own memory space, separate from the rest.

Does the Profile go in the global field?

There is a doubt worth clearing up, and the honest answer has two parts.

The official examples of what goes in that field are your preferred methods, the terms you use, typical scenarios and your general communication instructions. None of them mentions your job or your company. But that is an omission, not a prohibition: the documentation says neither "put your job here" nor "don't."

My call: put in the global field the stable professional slice that calibrates answers (your domain, your audience and your level), let Memory accumulate the rest and reserve projects for what is situational. The field is not for dumping personal history with no function. And since the documentation does not say whether Memory replaces or complements manual instructions, I use them as a complement.

How they combine with RICO and STEER

What	When you write it	What it is for
The 4Ps	Once	The base the tool carries into every conversation
RICO	In every prompt	A day-to-day task
STEER	In every prompt	An open task, where judgment matters more than speed

They work together. The 4Ps lay the base and, on top of it, you write RICO or STEER depending on the task. Since you already said in the 4Ps who you are and how you want answers, in the prompt you only need what is specific to that task.

Try it

Write the four Ps in four lines, in that order. Start with the Protocol: what it should do when information is missing.
Paste them into your tool's instructions field.
Run a RICO prompt you already used and compare the answer with the earlier one.
Review them when you switch tools or models. Old instructions can end up too prescriptive for newer models.