Give AI a Job Description, Not a Job Title: How to Assign Roles That Get Better Results

“Act as a world-class business strategist with 30 years of experience.”

It sounds impressive. The AI has apparently received a promotion, a corner office, and several decades of fictional employment history.

But does assigning an expert role actually improve an AI’s answer?

Sometimes. Just not for the reason many people assume.

A role can influence what an AI notices, which concerns it prioritizes, what language it uses, and how it organizes its response. What a role cannot do is install missing knowledge or guarantee that the answer is correct.

The useful part is not the imaginary résumé. It is the assignment.

An AI role is a decision filter, not a knowledge upgrade.

To get better results, stop searching for the most impressive job title. Give the AI a clear description of the work you need done.

What Is AI Role Prompting?

AI role prompting is the practice of assigning an AI a function or professional perspective that guides how it completes a task.

Consider three versions of the same request.

The second prompt gives the AI a title.

The third gives it a job description.

That distinction matters because labels such as “copywriter,” “financial expert,” or “business strategist” leave a great deal open to interpretation. A working role explains which parts of that profession are relevant to the task.

Current OpenAI guidance (opens in a new tab) similarly recommends stating the goal, relevant context, constraints, required evidence, success criteria, and desired output format. A role may help frame the work, but it does not replace those instructions.

Start With the Problem, Not the Profession

When choosing an AI role, do not ask:

Who would sound smartest doing this?

Ask:

What is preventing the current answer from being useful?

Choose the role based on that bottleneck.

The best role is not always the most prestigious professional.

Sometimes the most useful perspective is:

• A first-time customer who does not understand the terminology
• A busy reader who will leave if the answer takes too long to get to the point
• A cautious buyer comparing your product with a cheaper alternative
• A beginner trying to follow your instructions without prior knowledge
• Someone who disagrees with your conclusion

An expert may understand what you intended to say. A confused customer will notice that you never actually said it.

When using a customer or audience perspective, describe the relevant circumstances rather than relying only on a demographic label. For example:

That gives the AI more useful direction than simply naming an age group, occupation, or type of household.

What an AI Role Actually Changes

A useful role can change four important parts of an answer.

What the AI notices

A proofreader looks for spelling, grammar, and punctuation errors.

A developmental editor looks for structural problems, missing explanations, and sections that appear in the wrong order.

A skeptical customer looks for vague promises, unanswered questions, and reasons not to buy.

All three can examine the same page and produce very different findings.

What the AI prioritizes

A conversion copywriter may prioritize persuasion and momentum.

A compliance reviewer may prioritize precision, evidence, and cautious wording.

A plain-language editor may prioritize understanding over technical completeness.

None of these perspectives is automatically best. The right choice depends on the result you need.

How the AI explains something

Compare:

The second request guides the vocabulary, pacing, and depth of the explanation.

A role can also make an answer less suitable. Asking for an “elite academic expert” may produce more jargon and detail when the reader really needs a direct explanation in ordinary language.

How the answer is structured

Some roles naturally suggest useful deliverables.

An auditor might return:

• Finding
• Severity
• Evidence
• Recommended action

A project manager might return:

• Task
• Owner
• Dependency
• Deadline
• Risk

An editor might return:

• Problem
• Why it matters
• Suggested revision

This is why functional roles often outperform inflated identities. “Audit this process for failure points” gives clearer direction than “be a genius consultant.”

Why “Act as an Expert” Often Falls Short

The phrase is not useless. It is simply incomplete.

A peer-reviewed 2024 study (opens in a new tab) tested 162 personas across four families of language models and 2,410 factual questions. Adding personas did not improve overall performance compared with prompts that used no persona. The researchers also found that selecting the most effective persona for a given question was difficult to automate reliably.

A May 2026 preprint reached a similarly nuanced conclusion after comparing role-prompting methods across 1,140 open-ended questions and 38 expert roles. Expert roles tended to increase depth while reducing clarity, and their effects varied by task and subject area. The researchers concluded that persona prompting often reshapes response characteristics rather than broadly increasing capability.

In practical terms, an expert role may cause an answer to:

• Use more professional terminology
• Include additional caveats
• Discuss more specialized considerations
• Adopt a more authoritative tone
• Become longer and more structured

Those changes may be useful. They are not the same as becoming more accurate.

The AI can sound more like an expert without becoming more correct. Confidence has always owned a convincing jacket.

Instead of writing:

The second version replaces status with observable behavior.

Build a SoloPrompt Assignment

A useful role prompt does not need to be complicated. Define four things:

1. Assignment
2. Lens
3. Guardrails
4. Finish line

Together, these form a SoloPrompt Assignment.

1. Assignment

What is the AI responsible for noticing, deciding, creating, or improving?

The second instruction describes work that can be performed and evaluated.

2. Lens

Whose concerns or priorities should guide the work?

The role, lens, and audience are related, but they are not identical.

• Role: What function the AI performs
• Lens: Whose concerns it should consider
• Audience: Who will ultimately use the result

For example:

• Role: Line editor
• Lens: Skeptical first-time reader
• Audience: Small-business owners new to AI

3. Guardrails

What should the AI preserve, avoid, or refuse to assume?

Without guardrails, AI may “improve” an article by removing its personality, inventing convenient facts, or replacing the entire draft when you asked it to fix three paragraphs.

4. Finish line

What must a successful result contain?

“Give me a strong answer” is difficult to evaluate.

“Identify five problems and recommend a correction for each” gives the AI—and you—a visible finish line.

Here is the complete assignment:

The role selects a useful function. The assignment tells it what useful work to do.

Role, Task, Context, Standards, and Format Are Different Things

A common mistake is squeezing the entire prompt into one overstuffed role:

That is not really a role. It is a suitcase someone sat on to get the zipper closed.

Separate the prompt into distinct parts.

Role

Who is approaching the work?

SEO content strategist

Task

What must be accomplished?

Create an outline for an article about starting a service business with a small budget.

Context

What situation, audience, or source material matters?

The article is for employed adults considering a side business. They have limited time and are skeptical of exaggerated passive-income claims.

Standards

What qualities should guide the result?

Prioritize realistic startup costs, customer acquisition difficulty, recurring expenses, and time to first revenue. Avoid generic motivation and unsupported income claims.

Format

How should the answer be delivered?

Return a proposed title, search intent, introduction angle, and detailed H2/H3 outline with an example for each major section.

The role chooses the chair. The rest of the prompt explains why the meeting exists.

Weak AI Roles Versus Working Roles

Writing a blog post

Evaluating a business idea

At this point, the AI has six imaginary jobs and no clear assignment.

Editing an article

“Improve” could mean shorten it, expand it, formalize it, simplify it, or remove every detectable sign that a person wrote it.

Reviewing a sales page

The better prompts do not merely assign expertise. They define the decisions the role must make.

A Quick Role-Prompting Comparison

Suppose a product page contains only this copy:

The Budget Reset Planner makes budgeting simple. Download it today and take control of your money.

You could ask:

A reasonable answer might recommend adding benefits, proof, and a stronger call to action. None of that is wrong, but it is broad.

Now add a working role:

That version directs attention toward more specific weaknesses:

• “Makes budgeting simple” does not explain how the planner handles changing income.
• “Take control of your money” may sound like generic financial advice the reader has already heard.
• The copy gives no reason to believe this planner differs from budgeting systems the customer previously abandoned.

This is an illustration, not proof that a role always produces a better answer. The important point is that the role changed what counted as relevant.

Use Multiple Roles in Sequence, Not All at Once

Two compatible roles can sometimes work together:

Act as a content strategist and conversion copywriter.

But stacking many roles into one prompt often creates competing priorities. The AI tries to satisfy everyone and produces something resembling the minutes from a committee meeting.

For a complex project, use a role sequence:

1. Strategist: Defines the goal, audience, and approach.
2. Researcher: Collects or organizes the needed evidence.
3. Creator: Produces the first version.
4. Critic: Finds weak reasoning, missing details, and unsupported claims.
5. Audience reviewer: Tests whether the result is understandable and useful.
6. Editor: Applies the strongest revisions without losing the original purpose.

Breaking the work into stages makes each result easier to inspect. It also allows you to change the role when the nature of the task changes.

Anthropic’s current guidance includes explicit prompt chaining (opens in a new tab) as a useful approach when intermediate outputs need to be inspected or a particular workflow must be enforced. It also recommends defining success criteria (opens in a new tab) and evaluating prompts rather than assuming an instruction will work consistently.

When You Should Skip the Role Entirely

Not every request needs a character introduction.

A role often adds little when the task is mechanical, such as:

• Extracting dates from supplied text
• Reformatting content
• Alphabetizing a list
• Sorting information into defined categories
• Converting data into a specified structure
• Correcting capitalization according to a fixed rule

For example:

No imaginary forensic accountant is required.

Use a role when perspective, judgment, or prioritization changes what a good answer looks like.

Skip it when the task has objective rules and a clearly defined output.

Common Role-Prompting Mistakes

Giving the AI a role but no decision

You are a business consultant.

What should the consultant evaluate, compare, diagnose, or recommend?

A role without a decision is just a name badge.

Using a simulated persona as market research

AI can review a page through the lens of a cautious buyer or inexperienced user. That may help uncover questions and possible objections.

It does not replace interviews, surveys, analytics, or conversations with real customers.

Combining incompatible priorities

Be exhaustive, extremely brief, highly technical, easy for anyone to understand, formal, casual, and entertaining.

This is not a sophisticated prompt. It is a small workplace dispute.

Decide which qualities matter most and explain the tradeoff.

Keeping the same role after the work changes

A creative role may be useful during brainstorming and unreliable during verification.

A strategist may develop the plan. A researcher should check the claims. An editor should improve the presentation.

Switch roles when the job changes.

Treating polish as proof

A confident tone is not evidence.

For factual or high-stakes work, provide reliable sources, request citations, distinguish verified facts from assumptions, and independently check consequential claims.

A Reusable AI Role Prompt Template

Use this as a starting point:

Remove any line that does not improve the assignment. Longer prompts are not automatically better prompts. Clearer prompts usually are.

If you want somewhere to keep the versions that work, our personal prompt library template and guide to versioning and testing AI prompts pick up exactly where this template leaves off.

Test the Role Instead of Trusting It

A role should be treated as a testable part of the prompt, not a ritual phrase.

Run a simple comparison.

Prompt A

Give the AI the task, context, standards, and format—but no role.

Prompt B

Use the same prompt with a clearly defined working role.

Then compare the results:

• Did the role uncover more useful details?
• Did the response become more specific?
• Did it improve the organization?
• Did it introduce unnecessary jargon?
• Did it become less direct?
• Did it sound more confident without stronger support?
• Did it improve something you actually care about?

The useful question is not merely:

Did the role work?

It is:

What did the role change, and was that change helpful?

Keep the role when it improves the result. Revise or remove it when it does not.

Assign the Work, Not the Wardrobe

Assigning AI a role can improve an answer, but the improvement does not come from handing the model an imaginary résumé.

It comes from directing attention toward the right problem.

Instead of automatically typing “Act as an expert,” decide what is missing:

• Better judgment
• A different perspective
• Clearer priorities
• Stronger criticism
• More suitable language
• A better-defined output

Then give the AI an assignment, a useful lens, sensible guardrails, and a visible finish line.

When perspective matters, assign a role.

When the task is mechanical, skip it.

When the work is complex, use several focused roles in sequence rather than assembling the world’s most overqualified imaginary committee.

A title tells the AI who to pretend to be.

A job description tells it what useful work to do.

Turn your next task into a complete assignment →