Most people build role prompts ad hoc — they think of a job title, type it, and hope. That works for casual use and falls apart the moment you need personas that are consistent, reusable, and defensible across a team. What you need is a model: a repeatable structure that turns persona design from improvisation into a process anyone on your team can follow and review.
This article introduces PRISM, a five-stage model for designing role prompts: Purpose, Role, Inputs, Standards, and Maintenance. Each stage answers a specific question, and skipping any one of them is where most role prompts break down. The model is deliberately ordered — Purpose comes before Role for a reason, and Maintenance is a stage, not an afterthought. Work through it in sequence and you will produce personas that hold up under real use.
PRISM is not a rigid ceremony. For a quick task you might run the whole thing in your head in thirty seconds. For a production prompt serving thousands of requests, you might spend an hour on it. The value is having the same structure at both scales.
P — Purpose
Every persona starts with the output you are trying to produce, not the role.
The Question
What does a great output look like, exactly? Tone, length, structure, what it must include, what it must avoid.
Why It Comes First
You cannot choose a fitting role for an undefined target. Teams that start with the role end up with a persona that produces something — just not the thing they needed. Defining purpose first is the discipline our step-by-step approach to role prompting opens with.
R — Role
Now choose the persona whose natural disposition produces that output.
The Question
Whose voice, priorities, and habits naturally generate the output defined in Purpose?
Behavior Over Status
The role must describe behavior, not prestige. Not "a senior analyst" but "an analyst who leads with the bottom line, quantifies claims, and flags assumptions." The behavioral framing is what gives the model something to act on, as detailed in our best practices that actually work.
When to Use Multiple Roles
If the Purpose involves evaluation or multiple perspectives, assign several narrow roles across steps rather than one broad persona — the per-step pattern that consistently outperforms a single voice trying to do everything.
I — Inputs
A persona without context floats free. Inputs ground it in the specific situation.
The Question
What context and material does the role need to do the job?
The Two Kinds of Input
- Situational context: what the persona is working on ("revising a B2B onboarding email").
- Reference material: examples of good output, the source text, brand guidelines.
A role plus two or three examples vastly outperforms the role alone — the examples pin down the exact quality bar the persona can only gesture at.
S — Standards
Standards are the priorities and constraints that govern the output.
The Question
What does the persona optimize for, and what must it never do?
Priorities and Constraints
- Priorities: what wins when trade-offs arise ("clarity over completeness").
- Constraints: hard limits ("under 120 words, no jargon, no exclamation marks").
This is where you resolve tensions in advance. "Concise but thorough" is a contradiction; Standards forces you to pick. Skipping this stage produces the muddled output cataloged in our 7 common mistakes with role prompting.
M — Maintenance
The stage everyone forgets. A persona that works today decays without maintenance.
The Question
How will this role stay effective over time and across a conversation?
The Maintenance Practices
- Placement: persistent personas go in the system message to resist drift.
- Re-anchoring: in long conversations, restate the role when the voice starts fading.
- Re-testing: when inputs or goals change, re-run the comparison and confirm the role still earns its place.
Treating maintenance as a real stage is what separates a durable prompt library from a pile of stale one-liners. Run finished personas past our role prompting checklist for 2026.
Applying PRISM at Different Scales
The model flexes with the stakes.
Quick Tasks
Run P-R-I-S-M as a mental pass in under a minute: know your target, pick a behavioral role, supply context, set one priority and one constraint, place it correctly.
Production Prompts
Document each stage explicitly, test Standards against real inputs, and write down Maintenance rules so the next person can review and update the persona. The structure makes the prompt auditable, which matters the moment more than one person owns it.
PRISM Against a Real Task
Watching the model run on a concrete job shows how the stages interlock. Suppose you need a persona for drafting LinkedIn posts that summarize your company's research.
The Five Stages in Action
- Purpose: a 120-word post that opens with a counterintuitive finding, explains it plainly, and ends with a question that invites discussion — confident, not promotional.
- Role: a researcher-turned-communicator who leads with the surprising result and explains without jargon. Behavioral, not "thought leader."
- Inputs: the source research summary, plus two example posts in the voice you want.
- Standards: priority is the counterintuitive hook first; constraints are under 120 words, no hashtags, no hype words like "game-changing."
- Maintenance: persona in the system message; re-test when the research topic area shifts to confirm the hook-first structure still lands.
Each stage closed a gap the previous one left open. Purpose alone would not produce the voice; Role alone would float without Inputs; Standards alone would constrain an undefined target. The model's value is forcing all five.
Common Ways Teams Misuse the Model
PRISM helps only if applied honestly. A few patterns undermine it.
Skipping Straight to Role
The most frequent error is starting at R — picking a job title — and backfilling the rest. This reverses the model's logic and produces personas optimized for a role rather than for the output you needed. Always anchor on Purpose first, even when the role feels obvious.
Treating Standards as Optional
Teams under time pressure often drop the Standards stage, assuming the persona implies its own priorities. It does not. Without explicit priorities and constraints, the model picks its own, and contradictions go unresolved. Standards is where most of the precision lives; skipping it hollows out the rest.
Forgetting Maintenance Exists
Because Maintenance happens after the prompt ships, it is the easiest stage to neglect. Teams that treat PRISM as a one-time construction exercise end up with personas that worked at launch and quietly degraded since. Building Maintenance into the model as a named stage is the deliberate countermeasure to that drift.
Frequently Asked Questions
Why does Purpose come before Role in PRISM?
Because you cannot pick a fitting persona for an undefined target. If you choose the role first, you get output that is fine in the abstract but rarely the specific thing you needed. Defining the ideal output — tone, length, structure, inclusions, exclusions — first lets you select a role whose disposition actually produces it.
What makes the Role stage different from just naming a job title?
The Role stage demands behavior, not status. A job title like "senior analyst" carries almost no actionable signal. PRISM requires describing what the persona does — leads with the bottom line, quantifies claims, flags assumptions — because that is what the model can actually execute. Behavioral framing is the whole point of the stage.
Why is Maintenance a full stage rather than an afterthought?
Because personas decay. They drift over long conversations and go stale when goals change. Treating Maintenance as a real stage forces you to decide placement, re-anchoring, and re-testing up front. Teams that skip it end up with prompt libraries full of personas that quietly stopped working months ago.
Can I use PRISM for a one-off prompt?
Yes. For a quick task, run all five stages as a thirty-second mental pass — define the target, pick a behavioral role, supply context, set a priority and a constraint, and place it correctly. The same structure scales from a casual request to a documented production prompt; only the rigor changes.
How does the Inputs stage relate to using examples?
Inputs covers both situational context and reference material, and examples are the most powerful reference material you can provide. A role sets disposition, but two or three examples pin down the exact quality bar the persona can only gesture at. The Inputs stage is where you deliberately decide what examples to include.
Key Takeaways
- PRISM is a five-stage model: Purpose, Role, Inputs, Standards, and Maintenance.
- Purpose comes first because you cannot choose a fitting role for an undefined output.
- The Role stage requires behavioral description, not job titles or empty superlatives.
- Inputs grounds the persona in context and examples; Standards sets priorities and resolves contradictions.
- Maintenance is a real stage — handle placement, re-anchoring, and re-testing to keep personas from decaying.