Give It a Role to Play

Lesson 6 · taught by Pip

Here is the plain idea: when you tell the AI who to be before you tell it what to do, it answers differently. "Be a careful editor" and "be an excited copywriter" will treat the very same paragraph in two very different ways — and one of them is the one you actually wanted.

Think of it like calling the right shop in town. If you need your roof patched, you don't ring "someone." You ring the roofer. The moment you say the word, a whole bundle of judgment comes with it — what they check first, what they warn you about, the tone they take. A role does the same thing for the AI. One short phrase loads in a way of thinking.

Why a role does so much

The AI has read the work of editors, receptionists, teachers, and lawyers. When you name one, you point it at that body of habits. A "careful editor" starts noticing weak verbs and tightening sentences. A "friendly receptionist" starts being warm and asking what you need. You didn't have to spell out every rule — the role carried them in quietly.

It also sets the tone and the judgment at the same time. Ask a plain "fix this email" and you might get something stiff. Ask "you're a warm, busy office manager — fix this email so a customer feels looked after" and the whole feel shifts, because you told it whose shoes to stand in.

How to give a good one

State the role, then the job. A simple shape:

"You are a patient bookkeeper. Look over these expenses and flag anything that looks off."

You can add a little color to sharpen it — careful, friendly, blunt, cautious. Each word nudges the judgment. "Be a blunt editor" cuts harder than "be a gentle editor," and you get to pick which one suits the day.

A few honest cautions. A role shapes tone and focus — it does not make the AI an actual licensed expert. "Act as a lawyer" gives you lawyer-flavored wording, not legal advice you can bank on. And don't pile on five roles at once; one clear hat beats a costume closet.

This is also the quiet trick behind a business assistant. When Portmint builds a branded AI for a company, a big part of the work is giving it a steady role — this business's helpful front-desk voice, scoped to this business's world — so it never wanders off into being a general-purpose chatbot.

Your turn

Take a task you do often and put one role in front of it: "You are a [careful / friendly / blunt] [editor / planner / receptionist]." Run it once with the role and once without, and notice how the tone and the things it catches both change.

Next we'll circle back to context — the single biggest lever of all — and really settle it, because even the best role still needs to know your world. 🔦