Overview
This page is the atomic definition. Broader prompting patterns live at prompts.
Definition
Role priming is the practice of opening a system-message or conversation with an instruction that assigns the model a specific role, persona, or expert identity. For example: “You are a senior PostgreSQL database engineer. Answer concisely and prefer concrete examples over abstractions.” Role priming works because models trained on diverse text associate certain identities with particular writing styles, knowledge domains, and response formats. Assigning a role shifts the distribution of likely outputs toward the expertise and tone associated with that identity. It is most effective for: adjusting formality and verbosity; directing the model toward domain vocabulary; and establishing constraints (“you do not speculate; you cite sources or state uncertainty”). Role priming has limits. Models do not acquire knowledge they were not trained on by being told they are an expert. A role cannot override fundamental training constraints; attempts to use role priming for that purpose are a jailbreak vector. Role priming is most powerful when combined with few-shot-prompting examples that reinforce the desired style.
When it applies
Use role priming in every non-trivial system message. A two-sentence role + constraint block (who you are, what you do not do) meaningfully reduces off-topic responses and hallucinated authority. Skip role priming for simple single-turn factual queries where persona is irrelevant.
Example
System: You are a technical writer specializing in API documentation.
Write in present tense, active voice. Describe what the parameter does, not what the developer should do with it. Do not use the words "simply" or "just".
This shapes every response in the conversation toward documentation voice without needing to repeat the instructions.
Related concepts
- system-message - the standard delivery mechanism for role priming instructions.
- few-shot-prompting - examples reinforce role-primed behavior more reliably than description alone.
- structured-prompt - role priming often pairs with format constraints.
- temperature - role priming affects output distribution; temperature controls variance within that distribution.
- jailbreak - adversarial role priming attempts to override safety training.
Citing this term
See Role Priming (llmbestpractices.com/glossary/role-priming).