Artificial Idea | AI careers · practical prompts · no hype Thursday, August 21, 2025 · Issue #6 · Prompt Tutorial
You have been talking to AI like it is a search engine. Here is how to talk to it like it is the smartest person in the room.
Monday's profession-by-profession breakdown generated more replies than anything we have published so far. The theme running through most of them was the same: people understand the map now, but they are not sure how to start moving on it. What does "engage with AI tools seriously" actually look like on a Tuesday afternoon, inside a real job, with real deadlines?
This issue answers that question with the most powerful single technique in practical prompt engineering, one that costs nothing to learn, takes thirty seconds to apply, and produces a quality difference in AI output that most people find genuinely surprising the first time they see it.
It is called role assignment. And if the 4-part framework from Issue #4 is the foundation of good prompting, role assignment is the load-bearing wall.
What role assignment actually does
When you open ChatGPT or Claude and type a question cold, no context, no setup, no instructions, the model responds as a generalised assistant calibrated for the average user asking the average question. It is helpful in a broad, shallow way. It covers the basics. It hedges appropriately. It does not offend anyone.
It also does not particularly impress anyone.
Role assignment changes the operating parameters of the conversation before it starts. When you tell a model "you are a senior strategy consultant with fifteen years of experience advising consumer brands on market entry", and you mean it, and you write the rest of the prompt accordingly, you are not engaging in a creative exercise. You are activating a different part of the model's learned behaviour. The vocabulary shifts. The assumptions change. The depth of the response increases. The hedging decreases in proportion to the specificity of the role.
This happens because language models learn from text produced by real people in real contexts. A senior strategy consultant writes differently than a generalist. They ask different clarifying questions. They structure their analysis differently. They flag different risks. The model has seen enormous quantities of text from people operating in that role, and when you explicitly invoke the role, you pull that pattern forward into your conversation.
The result is not a perfect simulation of a strategy consultant. It is a response that is meaningfully more useful than what you would get without the instruction, and in most professional contexts, meaningfully more useful is exactly what you need.
The difference in practice
Here is the same request, sent two ways.
Without role assignment: "What should I consider when expanding my business into a new market?"
The output you get will be a competent, general list. Define your target customer. Research the competitive landscape. Understand regulatory requirements. Consider localisation. It is not wrong. It is also not particularly useful to someone who already knows their industry and needs sharp, specific thinking, not a checklist that could apply to any business in any market anywhere in the world.
With role assignment:
You are a senior market entry strategist who has spent fifteen years
advising mid-sized B2B technology companies expanding into Southeast
Asian markets. You have seen expansions succeed and fail, and you
have strong opinions about where companies underestimate the
complexity.
I run a B2B SaaS company with 200 employees and $8M ARR, currently
operating only in India. We are considering expanding into Singapore
and Indonesia in the next 18 months.
What are the three most common mistakes companies like mine make at
this stage, and what would you want to know about our situation
before giving a more specific recommendation?
The output from the second prompt is categorically different. It is specific to B2B technology. It reflects the dynamics of Southeast Asian market entry. It leads with failure modes rather than a generic checklist. And the instruction to ask clarifying questions before giving recommendations produces a back-and-forth that surfaces the actual nuances of your situation, which is where the genuinely useful advice lives.
Same underlying question. Completely different quality of engagement.
Five role assignments worth keeping
These are broadly applicable across professional contexts. Save them, adapt the specifics to your industry, and use them as starting points rather than rigid templates.
The strategic advisor:
You are a senior strategy consultant with deep experience in
[your industry]. You are direct, you challenge assumptions, and
you are more interested in asking the right questions than
giving comfortable answers. You have seen enough failed strategies
to be appropriately skeptical of plans that sound good on paper.
Best for: Business decisions, strategic planning, evaluating options, stress-testing ideas before you present them.
The experienced hiring manager:
You are a senior hiring manager with fifteen years of experience
recruiting for [your function] roles at competitive companies.
You have reviewed thousands of CVs and conducted hundreds of
interviews. You know exactly what strong candidates look like
and you are not easily impressed.
Best for: CV review, interview preparation, evaluating your own professional positioning, understanding what interviewers are actually looking for.
The plain-language editor:
You are a professional editor whose speciality is taking complex,
jargon-heavy professional writing and making it clear, direct,
and accessible without losing accuracy. You have a low tolerance
for corporate language, passive voice, and sentences that say
nothing. Your job is to make every sentence earn its place.
Best for: Editing reports, proposals, presentations, emails, anything where clarity matters and you have drifted into corporate autopilot.
The devil's advocate:
You are a highly intelligent, well-informed critic whose job
is to find the weaknesses in any argument or plan I present.
You are not contrarian for the sake of it, you are genuinely
trying to find the holes before someone else does. You do not
soften your critique. You lead with the strongest objection first.
Best for: Testing business plans, proposals, recommendations, any piece of work you are about to stake your reputation on.
The domain expert:
You are a [specific expert, tax lawyer / UX researcher /
supply chain specialist / clinical psychologist] with twenty
years of experience. You are speaking to a professional peer
who has general business knowledge but is not a specialist
in your field. You can use technical language where necessary
but always explain why it matters practically.
Best for: Getting specialist perspective on problems outside your core expertise without needing to hire a consultant for every question.
The one thing that makes role assignment fail
Role assignment works in proportion to the specificity and commitment of the rest of the prompt. If you assign a detailed, credible role and then ask a vague, one-sentence question, the role instruction cannot save the output. The model needs something to work with.
The formula that consistently produces the best results: detailed role + sufficient context + specific task + clear constraints. The role assignment is the first layer of that stack. It is not a substitute for the others.
One more thing. After you assign a role, commit to treating the conversation as if the role is real. Ask follow-up questions. Push back on the output. Ask the "consultant" to elaborate on the point you found most useful. Ask the "editor" why they changed a specific sentence. The back-and-forth is where the real value emerges, and it only happens if you engage with the role you set up rather than treating the first response as the final answer.
This week's prompt to try
Take the most difficult professional decision or problem currently sitting on your desk. Write a role assignment for the type of expert you would most want in the room for that conversation. Use the structure from Issue #4, role, context, task, constraints. Run it.
Then ask one follow-up question based on the response. Notice what happens to the quality of the conversation when you treat it as a dialogue rather than a query.
Monday we are looking at something that does not get enough serious coverage: what the WEF's Future of Jobs report actually says when you read the full document rather than the press release. The gap between the two is significant, and some of what is buried in the methodology section changes the picture considerably.
The headline version of that report has been cited in approximately ten thousand LinkedIn posts. Almost none of them read past page four.
We read the whole thing.
— The Artificial Idea team
