Why Telling AI to Be Shakespeare-Ogilvy Is Making Your Prompts Worse
The role prompt was a useful shorthand. It also has its limits.
This is common advice: Give the AI a role before you give it a task.
“You are a JPMorgan stock analyst. Evaluate this company.”
“You are a Leo Burnett-trained copywriter. Write me a landing page.”
“You are a senior product manager at Google. Review this strategy document.”
The appeal is clear: a role signals expertise and expectations, making it easy to request specific output without lengthy explanation. This shortcut works in straightforward cases, but breaks down when the role no longer aligns with reality.
The problem appears when the role stops mapping onto anything real.
“You are Shakespeare and David Ogilvy combined. Write a poem about my SaaS company.”
“You are the God of Weather Forecasting and a veteran commodity trader. Predict the corn market for September.”
“You are a top prompt engineer in Anthropic with 20 years of experience.”
These prompts ask the model to inhabit a persona that is incoherent (two writers with fundamentally different styles fused into one), fictional (there is no body of knowledge for what a weather deity says about corn futures), or self-contradictory (prompt engineering as a discipline did not exist 20 years ago).
The model cannot draw on real expertise because none exists for the role. So it improvises. And improvisation dressed up as authority is where AI output starts to sound confident and hollow at the same time.
But remember, the role was always just a proxy.
Here is the issue: even when role prompts work, they do so for reasons unrelated to the role itself.
When a “JPMorgan analyst” produces useful output, it works because the label efficiently packages a domain (finance), a register (professional and precise), and an implicit audience (someone who understands markets). The role was doing the work of context in shorthand. The model was not actually channelling a JPMorgan analyst. It was processing a compact signal that said: respond like a financial analysis written for someone who knows the field.
The shorthand is fine when the role is grounded, and the task is simple. It collapses the moment either condition fails.
The shift that fixes it
If the role is a proxy for context, the better move is to skip the proxy and give the context directly.
Instead of: “You are a JPMorgan analyst. Evaluate Apple.”
Try: “I am evaluating Apple as a position trade with a six-month holding period. I want an analysis focused on institutional flow signals, earnings momentum, and competitive positioning against Samsung and Google in the AI race. Assume I understand standard financial metrics. Skip the basics. Flag any contrarian indicators you spot in the data.”
That brief contains everything the role implied, plus specifics the role could never carry. The model does not have to guess what kind of analyst you want or what you already know. You have told it exactly what you need.
The output that comes back is dramatically more useful every time.
Why this matters now
Role prompting seems efficient, but its shortcuts often result in generic, less useful AI output.
Strong AI output comes from clear context, not cleverer roles.
This shifts prompting from theatrics to function. The key is briefing AI clearly, not role-playing.
Coming next
The next article in this series breaks down what a complete brief looks like. The four elements every effective prompt needs, and how to apply them, whether you are writing an email, analysing a stock, or briefing your team.
PromptCraft v2 takes this further, into a full system for working with AI across any tool. If you want to be the first to know when it is out, subscribe to The Intelligent Playbook now and get notified about its release as soon as it’s available.
And stop telling AI to be Shakespeare. Instead, start now: write a brief for your next AI task that clearly states what you need. See the difference yourself.