“You are a helpful assistant.” Most people who include a role instruction in their AI brief write something close to this. They have heard that telling AI to play a role improves the output, and so they add a line before their request. The output they receive is essentially identical to the output they would have received without it. A role instruction that describes “a helpful assistant” is not a role instruction. It is the default. It describes what AI already is. Writing it changes nothing. A role instruction that works does something different: it calibrates the AI’s perspective, expertise level, operating standards, and assumptions before any task begins. The difference between a role instruction that works and one that doesn’t is not whether it is present — it is whether it contains actual calibration information.
What a Role Instruction Is Actually For
AI approaches every task with a set of default assumptions about who it is, what level of expertise it is bringing, and what standards apply to its output. Without a role instruction, those defaults are calibrated to the broadest possible audience: a general-purpose assistant producing output that is accessible to an educated non-specialist. For most professional and specialist tasks, that calibration is wrong. A researcher needs a collaborator calibrated to academic standards, not general accessibility. A lawyer needs analysis that applies legal reasoning at a professional level, not a summary any reader could follow. A senior executive needs strategic output that assumes business sophistication, not output that explains what a P&L is. The role instruction is the mechanism for changing the default calibration. It tells AI who it is in this specific task — and therefore what assumptions to make, what level of rigor to apply, what to take for granted and what to explain.
The Three Things a Working Role Instruction Contains
A role instruction that changes AI output contains three elements: the domain and level of expertise, the operating context, and the implicit standards that expertise applies. The domain and level: not “a marketing expert” but “a senior brand strategist with experience in B2C consumer goods, particularly in category entries where established brands have incumbency advantages.” The specificity of the expertise changes what the AI treats as obvious, what frameworks it applies, and what depth of analysis it produces. The operating context: what situation the role is operating in. “Advising a founder” produces different output than “advising a board.” “Writing for a peer-review journal” produces different output than “writing for an industry publication.” The context calibrates the register, the assumed knowledge level of the audience, and the standards for what counts as sufficient. The implicit standards: the things that the role would never do, the quality thresholds that define the role’s professional standards, the ways of thinking that the expertise requires. “Do not make recommendations without acknowledging the limits of the available information” is an implicit standard. “Treat contested empirical claims as contested, not settled” is another. These are the professional habits that distinguish real expertise from the simulation of it.
What a Working Role Instruction Looks Like
Not working: “You are a financial advisor.” Working:
Role: You are a fee-only financial planner advising a dual-income household
in their mid-thirties with a combined income of approximately $180k, no
existing investment portfolio, and two competing priorities: building an
emergency fund and beginning retirement contributions. You do not sell
products. Your advice is calibrated to long-term outcomes, not short-term
comfort. You acknowledge uncertainty where it exists and do not present
projections as more reliable than they are.The first version adds nothing. The second changes the AI’s perspective, its standards of evidence, its awareness of the conflict of interest it does not have, and its calibration to a specific financial situation. The output from the second is categorically different.
The Role Instruction as a Transfer of Professional Standards
The most valuable thing a role instruction can do is transfer professional standards — the implicit habits of high-quality work in a specific field — into the brief. A research role instruction that includes “treat methodological distinctions as meaningful, not cosmetic” transfers a standard that separates careful academic work from popular science writing. A legal role instruction that includes “flag where the analysis depends on jurisdiction-specific rules that may not apply” transfers a standard that distinguishes rigorous legal analysis from confident-sounding generalization. These standards are what differentiate expert output from competent generalist output. They do not require long instructions — they require the right ones. A single sentence that establishes a professional standard can change the quality of everything that follows. For professionals who use AI regularly in complex domains, Briefing Fox helps structure the role instruction as part of the briefing process — ensuring the expertise calibration is built in before the task begins.
Before Your Next AI Task
Before any AI task that requires expertise, write a role instruction that includes three things: the specific domain and level of expertise, the operating context, and one professional standard that should govern the output. Skip “helpful assistant.” Write who you actually need. The role instruction is a two-sentence investment that changes everything that follows. Try Briefing Fox free at www.briefingfox.com.