Role and Persona Prompting: Making AI Sound Like Someone Specific, Part 2

Setting the role primes the model's vocabulary, focus, and standards. The same task — 'review this code' — gets a totally different answer when prefixed with 'you are a security engineer' vs. 'you are a junior dev'.

Some examples

• 'You are a senior backend engineer' makes Claude flag missing transactions you'd otherwise miss.
• 'You are a kindergarten teacher' makes ChatGPT explain photosynthesis with picture analogies.
• 'You are an editor for The Atlantic' tightens prose better than 'edit this'.
• 'You are an SRE on call' makes Cursor focus on log lines and timeouts, not refactors.

Try it!

Take a prompt you use often. Run it with three different roles up top. Compare the answers.

Models follow 'don't' instructions better than they used to, but not perfectly. Stack a few clear negatives ('no preamble, no apologies, no emojis') and you'll get the cleaner output you actually wanted.

Some examples

• 'Don't apologize or preamble' makes Claude open with the answer instead of 'Certainly!'.
• 'No emojis' on a serious doc actually keeps GPT-4 from sneaking in a sparkle.
• 'Don't add disclaimers' shortens medical-info answers to the actual content.
• 'No markdown headers' keeps ChatGPT producing flat text suitable for a plain-text email.

Try it!

Add a 'do not' clause to a prompt where the AI keeps doing something annoying. Note compliance rate.

role assignment activates patterns the model already knows

Some examples

• You are a senior accessibility reviewer
• You are a punk-rock zine editor
• You are a 1990s sysadmin

Try it!

Open your favorite AI tool and try one of the examples above. Pick the one that matches what you are actually working on this week. Spend 10 minutes, no more. Notice what worked and what did not — that's the real lesson.

reusable prompts compound; one-off prompts evaporate

Some examples

• A snippets file in your notes app
• Variables you fill in each time
• A changelog when you tweak one

Try it!

Open your favorite AI tool and try one of the examples above. Pick the one that matches what you are actually working on this week. Spend 10 minutes, no more. Notice what worked and what did not — that's the real lesson.

Role prompts set tone and vocab but don't actually unlock secret expert knowledge.

Some examples

• Use roles for tone ('explain like a friendly TA').
• Don't rely on roles for accuracy — use specific instructions.
• Combine: role for vibe + clear task for substance.

Try it!

Take a prompt with just a role. Add specific task instructions. Compare quality.

Understanding "Role prompts: 'you are a senior X' is mostly placebo" in practice: Prompting is a skill: the more specific and structured your input, the more useful the output. Telling AI 'you are a senior dev' helps a little — but specific instructions help a lot more — and knowing how to apply this gives you a concrete advantage.

• Use role, context, task, and format in every prompt
• Iterate: treat first outputs as drafts, not finals
• Use few-shot examples for complex formatting tasks
• Test prompts at different temperatures for creative vs. factual tasks

1. 1Rewrite one of your best prompts using role + context + task + format
2. 2Ask an AI to critique your prompt and suggest improvements
3. 3Compare outputs from two models using the same prompt

Most people prompt like: 'help me write an essay'. The pros prompt: '[Role] You're a high school English teacher. [Task] Help me write a 5-paragraph essay on The Great Gatsby's symbolism. [Context] I'm a 10th grader and need to focus on the green light. [Format] Give me an outline with 3 supporting quotes from the book.' Same AI, 10x better answer. The formula is Role-Task-Context-Format.

Some examples

• Role: 'You are a [type of expert]'
• Task: 'Help me [exact thing]'
• Context: 'I'm [your situation], the goal is [what success looks like]'
• Format: 'Output as [bullets / outline / 200 words / table]'

Try it!

Take your last AI prompt. Rewrite it using all 4 parts. Run both. The new answer will be visibly better. That's the formula.

Models have studied every job in the world. Telling them to play a specific role triggers the right vocabulary, examples, and rigor. The same question with no role gets the average; with a role gets the expert.

Some examples

• Ask Claude to be your AP Bio tutor for the next 10 messages and notice the depth difference.
• Ask ChatGPT 'role: hostile college admissions officer' and watch your essay get destroyed (helpfully).
• Ask Gemini to play three characters in sequence to triangulate an answer.
• Ask Perplexity for the role that makes it cite the most sources by default.

Try it!

Take your next chat. Open with 'You are an experienced [thing].' Compare to a no-role version of the same question.

The same five prompts handle 80% of your homework, social, and career writing. Saving them once turns AI from improv to a system. Adults call this prompt engineering; you can just call it not retyping.

Some examples

• Ask Claude to draft a starter prompt library of 10 prompts every teen should save.
• Ask ChatGPT for the Notion template most teens use for their prompt library in 2026.
• Ask Gemini to organize your prompts by category: school, social, career, fun.
• Ask Perplexity for the public prompt libraries shared by other students.

Try it!

Open Apple Notes. Create one note called 'Prompts.' Save the next prompt that works for you tonight. You started a library.

Every AI conversation has a hidden 'system prompt' that shapes how the model behaves throughout. When you set one yourself — in custom GPTs, Claude Projects, or API calls — you control the AI's persona, expertise, format, and limits across every reply. Mastering system prompts is the single biggest jump in prompting skill.

Some examples

• A system prompt defines role, audience, format, and constraints.
• Be specific: 'You are a tutor for high school AP Bio. Use Socratic questioning. Never just give answers.'
• Add a 'don'ts' list: 'Never recommend medical action. Never write essays for the user.'
• Test variations — small wording changes have big effects.

Try it!

Write a system prompt for an AI study partner. Use it for one week and refine based on what's missing.