Skip to main content

The "Act As" Hack is Overrated: Do This Instead - Digital Compliance Academy

Everyone tells you to start prompts with "Act as an expert..." But that is lazy prompting. Here is how to set real context that actually improves output.

Jon McGreevy August 23, 2025 3 min read
Prompt Engineering Context CLEAR Framework Advanced Skills

If I had a pound for every LinkedIn influencer who told me the secret to AI is saying “Act as a world-class copywriter…” I would own a private island by now.

It is the most common piece of advice in prompt engineering. And it is arguably the laziest.

Telling Claude or ChatGPT to “Act as an expert” provides a persona, but it doesn’t provide context. And context is king.

If you say “Act as a lawyer,” the AI will start using legal jargon. But it doesn’t know what kind of lawyer, for who, in what jurisdiction. It might give you a US District Court opinion when you need UK Employment Law advice.

The Problem with “Act As…”

  1. Stereotypes: It relies on the AI’s training data stereotype of that role. “Act as a Salesman” usually results in cheesy, aggressive “Glengarry Glen Ross” copy, because that’s what the internet thinks a salesman sounds like.
  2. Lack of Constraints: An “Expert” can still be verbose, wrong, or boring.
  3. Missing Goal: Who is the expert talking to? A physics expert talks differently to a PhD student than to a 5-year-old.

Context > Persona (The W-W-W Formula)

Instead of just assigning a role, you need to build a Simulation.

At DCA, we teach the W-W-W Formula to replace “Act As”:

  1. Who are they? (Role + seniority + attitude)
  2. Where do they work? (Industry + company culture)
  3. What is the specific challenge? (The “Why” behind the task)

Example 1: HR Policy

  • Lazy: “Act as an HR Manager. Write a remote work policy.”
  • W-W-W: “You are a pragmatic HR Director (Who) at a fast-growing tech startup in London (Where). We have a culture of high trust but low documentation. You need to formalise our remote work rules without sounding corporate or bureaucratic (What).”

Example 2: Sales Email

  • Lazy: “Act as a Copywriter. Write an email.”
  • W-W-W: “You are a Direct Response Marketer (Who) selling high-end consultancy services (Where). You are writing to cold prospects who are sceptical of sales emails (What). The goal is to get a reply, not a click.”

Why this works better in Claude vs ChatGPT

  • Claude 3.5 Sonnet excels at “System Prompting” (adopting a persona). If you give it the deep W-W-W context, it stays in character for the entire chat. It picks up the nuance of “fast-growing startup” and adjusts its tone to be less stuffy.
  • ChatGPT (GPT-4o) tends to drift. It starts with the persona but reverts to “Assistant” mode after a few turns. You need to remind it: “Remember, you are the sceptical HR Director.”

The “Anti-Persona” (Negative Constraints)

Sometimes, it’s more effective to tell the AI who not to be.

“Act as a marketing expert, but DO NOT sound like a LinkedIn influencer. Do not use buzzwords like ‘unlock’, ‘unleash’, or ‘supercharge’.”

This “Negative Persona” forces the model to drop its default, safe, “marketing speak” patterns and write something authentic.

So stop playing dress-up with “Act as,” and start building high-fidelity simulations. The AI is an actor. Give it a script, not just a costume.