Prompting
Role Prompting
A prompting technique that asks the model to respond as if it were a specific kind of expert, persona, or professional role.
Definition
Role prompting is the practice of beginning a prompt with an assigned identity — 'You are a senior copy editor,' 'Act as a UX researcher reviewing this flow,' 'Respond as a skeptical investor.' The mechanism is not that the model becomes the role; it is that the role conditions the probability distribution over which tokens are likely to come next. Outputs cluster around the vocabulary, structure, and concerns that role typically uses. Role prompting is one of the cheapest, most reliable ways to shift output style without writing detailed format instructions.
Why it matters
Used well, role prompting compresses a lot of implicit instruction into a few words. 'You are a senior product writer at a B2B SaaS company' carries assumptions about register, audience, and what counts as good — assumptions that would take a paragraph to spell out. Used badly, role prompting is decorative. 'You are a world-class expert' adds nothing because every model has seen the phrase a million times and it does not condition output toward anything specific. The practical test is whether the role implies a different distribution of vocabulary and structure than the default. 'A senior editor at The Economist' does. 'A 10x AI prompt engineer ninja' does not — it sits in a cluster of similar phrases the model has been trained to recognise as decorative and not informative.
Examples
Effective role
'You are a B2B SaaS landing-page copywriter who has shipped pages for Linear, Notion, and Vercel.' The role conditions tone, length, and structural expectations.
Ineffective role
'You are the world's best writer.' Decorative — implies no specific distribution and does not shift output meaningfully.
Role plus constraint
'You are a copy editor with a strict word limit. Cut this paragraph to 40 words while preserving the argument.' Role plus constraint outperforms either alone.
Frequently asked
Does role prompting actually work or is it superstition?
Both. A specific, plausible role meaningfully shifts output; a generic 'expert' role does not. Test on your own task to see which side of the line a role falls on.
Should I put roles in the system prompt or the user message?
Either works. System prompts are more persistent across multi-turn conversations; user-message roles are more flexible per task.
Can I stack roles?
Sparingly. 'A copy editor and a fact-checker' is fine. Long chains of roles dilute the signal and the model picks the closest match anyway.
Related terms
Prompt Engineering
The discipline of crafting model inputs — instructions, context, examples, and constraints — to produce reliably better outputs.
System Prompt
A persistent set of instructions that frames a model's behaviour across an entire interaction or product.
Few-Shot Prompting
Showing a model two to five worked examples of the desired input-output pattern before asking it to handle a new case.
Tone
The emotional register and stance a piece of writing communicates, independent of its literal content.
Put the concept to work
Open the rewrite engine and apply this principle to a draft of your own.
Try a rewrite