Most people open a prompt with what they want done: “Summarize this contract.” The model complies — generically. It has no idea whether you want a paralegal’s eye for liability or a founder’s eye for deal terms, so it splits the difference and gives you something bland.
The fastest fix costs you one sentence at the top: tell the model who it is before you tell it what to do.
You are a senior contracts attorney. Review this agreement and flag the three clauses most likely to hurt a small business.
Why a role changes everything
A role isn’t roleplay — it’s a shortcut to a whole set of defaults. Naming one quietly sets the model’s vocabulary, the depth it goes to, the assumptions it makes, and the things it thinks to check without being told. “Senior attorney” pulls in liability, indemnification, and termination. “Growth marketer” pulls in funnels, CAC, and hooks. You get the expert’s instincts, not just the expert’s words.
Before and after
Watch what the role does to an otherwise identical request:
❌ Explain what this error means and how to fix it.
✅ You are a senior TypeScript engineer doing a code review.
Explain what this error means, the most likely root cause,
and the smallest safe fix. Show the corrected snippet.Same question. The second version gets you a root cause, a minimal diff, and working code — because that’s how a senior reviewer answers by default.
Three roles worth keeping on hand
- The critic — “You are a skeptical editor. Find the three weakest points in this argument.” Great for pressure-testing your own writing or plans.
- The teacher — “You are a patient tutor. Explain this so a smart beginner gets it, then check my understanding with one question.” Turns answers into actual learning.
- The operator — “You are a pragmatic ops lead. Give me the shortest path to done, with the risks called out.” Cuts theory and gets you moving.
How Amplify does this for you
Picking the right role for every prompt is exactly the kind of thing Amplify handles automatically. It reads what you’re trying to do, chooses a role that fits the task and the model you’re sending it to, and layers it into a cleaner rewrite — before you hit send. You get the expert framing without having to remember to write it.
Try it now: take your last so-so prompt, add one role sentence to the front, and re-run it. The difference is usually obvious on the first try.