Role and persona
State which details the model should look at, and for which reader.
- use
- levels
- simple · medium · hard
- links
- 3 related methods
- license
- CC BY 4.0 · View the card’s open source
What is it?
Giving a role means choosing the angle from which a text is read. When you say “editor for a beginner reader”, you can ask it to focus on unclear terms; when you say “developer reviewing the tests”, on boundary conditions. Explaining the role's job with concrete verbs is easier to check than writing a job title alone.
An expert persona does not give the model new knowledge, a diploma or tool access. Having three characters speak inside one response is also not three independent agent calls.
When does it help?
Use it when you edit the same text for different readers or choose a specific review goal. Also give the object the role should look at and the change it should deliver.
Examples
The situations and responses below are fictional teaching examples; they are not results from a model, tool or benchmark that was actually run.
Simple
Situation
New participants do not understand the phrase “registration verification”.
Prompt
Prompt
Review this as an editor for a beginner-level reader. Sentence: “Subsequent to registration verification, your participation will be activated.” Meaning: the place is confirmed when the confirmation email arrives. Simplify the formal words; write one sentence without adding a new condition.Sample output
Your place is confirmed when the confirmation email arrives.
What did we get?
The role narrowed down which language problem to look at. Check separately that the sentence describes the registration process correctly.
Medium
Situation
You want to read the shipping function separately from an interface-text angle and a testing angle.
Prompt
Prompt
Use two review angles within a single response; they are not independent agents.
Information: “Free shipping for 500 TL and above.” Code: total > 500 ? 0 : 50.
Editor: state in one sentence the threshold the text promises.
Test reviewer: propose one boundary test for this promise and its expected result.
Do not claim that you ran the code.Sample output
Editor: Exactly 500 TL falls in the free range. Test reviewer: for total=500 the expected shipping is 0; the current expression produces 50.
What did we get?
The two roles produced different review questions. Both are parts of the same response; the real test is the developer's job.
Hard
Situation
In a decision note, cost and accessibility conflict. You do not want the role to take over the decision.
Prompt
Prompt
You are the editor of a decision note; do not make the choice for me.
Venue A: 800 TL, step-free entry verified. B: 600 TL, three steps at the entrance. For an 18-person team, step-free access is mandatory. The budget ceiling is 700 TL.
First separate the mandatory conditions. Then state which condition each option fails. Do not assume a budget change that is mine to make. Write at most the next decision question.Sample output
A exceeds the budget by 100 TL; B does not meet the access condition. With the current information, there is no suitable option. Should we look for a third venue that is step-free and costs at most 700 TL?
What did we get?
The role organized the note; it did not create new budget authority. The decision to start a search or change the scope stayed with you.
Where should you stop?
Do not treat adjectives such as “the world's best expert” as a guarantee of factual accuracy. The results of persona studies depend on the task and the model. CO-STAR organizes brief fields such as context, objective and reader together instead of a role; multi-agent methods also need a real call infrastructure.
Sources
A Prompt Pattern Catalog to Enhance Prompt Engineering with ChatGPT (new tab) — White, Jules; Fu, Quchen; Hays, Sam; Sandborn, Michael; Olea, Carlos; Gilbert, Henry; Elnashar, Ashraf; Spencer-Smith, Jesse; Schmidt, Douglas C.. 2023-02-21. Defines the persona pattern as a form of interaction; it is not a certificate of professional competence. Evidence level: relevant body sections of the original paper.
When “A Helpful Assistant” Is Not Really Helpful: Personas in System Prompts Do Not Improve Performances of Large Language Models (new tab) — Zheng, Mingqian; Pei, Jiaxin; Logeswaran, Lajanugen; Lee, Moontae; Jurgens, David. 2023-11-16; version read 2024-10-09. Examines how, in the models studied, adding a persona did not bring consistent improvement on factual tasks. Evidence level: relevant body sections of the original paper.
[2512.05858] Prompting Science Report 4: Playing Pretend: Expert Personas Don't Improve Factual Accuracy (new tab) — Basil, Savir; Shapiro, Ina; Shapiro, Dan; Mollick, Ethan; Mollick, Lilach; Meincke, Lennart. 2025-12-05. Provides an abstract-level limit on the relationship between expert personas and accuracy; the full experimental body was not read for this record. Evidence level: abstract only.
How this image was made
Original illustration made with Google Gemini · 1024 × 572
Generate an image: Create an original museum-grade computational sculpture photographed as a physically present installation, horizontal 16:9. Deep anthracite void, mineral porcelain whites, restrained ice-blue and amber accents, fine particles only where structurally meaningful, tactile surfaces, subtle volumetric illumination, believable depth, exceptional edge detail and intentional negative space. A single complex white mineral object seen through a suspended amber optical lens that isolates its surface fissures while the rest remains visible in cool shadow. No human eye. A photographic lateral composition, one lens, one object. Keep the image sophisticated and legible at mobile size: one dominant mechanism, a clear silhouette, no decorative overload. No text, letters, numbers, typography, captions, symbols, watermark, logo, fake chart, screenshot, interface, identifiable people or artist signature. Do not imitate any specific existing artwork.