Prompt Library by Role — Free Online Tool on Toolpile

Pricing

Every tool you need, one place

PDF, Image, AI, Dev, and Business tools — free, private, no signup.

PDF

20

Image & Media

17

AI & Prompts

21

Text

13

Developer

26

Converters

6

SEO & Marketing

8

Generators

12

Productivity

7

Calculators

4

Business

9

Privacy & Security

5

Design

4

Content

5

Education

4

About Prompt Library by Role

A library of prompts is only useful if the prompts inside it have actually been used by someone who knew what they were doing. This one is organized by role — pick yours, copy what fits, modify the rest. No fluff, no AI-generated filler, no "act as a world-class expert" openers.

Who this is for

If you do one of six jobs — PM, software engineer, marketer, teacher, designer, or founder — this library is sized for you. Each role has 6 to 10 prompts that solve a real job-to-be-done in that workflow: write a PRD skeleton, run a senior-level code review, sharpen a positioning statement, build a lesson plan, audit a design for accessibility, prep an investor update. They're written assuming you know your domain — they don't pad with definitions or boilerplate. Drop them into Claude, ChatGPT, or Gemini and they work.

The prompts are hand-written by people who do these jobs, not generated by another LLM. That matters more than it sounds: AI-generated prompt libraries quietly drift into vague, self-referential language ("Act as a 10x engineer and provide insightful guidance…") because the model that wrote them was optimizing for plausibility, not utility. Every prompt here was authored against a specific task that a human had actually struggled with — usually because the obvious zero-shot prompt for that task produced mediocre output.

How to use this library

Each prompt is structured the same way: a clear role, a specific task, an output format, and constraints that prevent the model from hedging or wandering. To use one:

  1. Pick your role from the pills at the top. The library defaults to Product Manager because that's the most common visitor; click any other role to swap.
  2. Browse the cards. Each shows the prompt's purpose, a one-line preview, the recommended model (Claude, GPT-5, or Gemini — based on which model the author found gave the best output), and a tag.
  3. Click a card to open the full prompt. Read it first — these aren't black boxes. The structure (role → task → format → constraints) is the same one you can use to write your own.
  4. Replace placeholders in double curly braces (e.g. {{idea}}, {{topic}}) with your actual context. Don't skip this — the prompts assume you'll give them real input. Sending them with placeholders intact produces generic output.
  5. Copy the prompt and paste it as the first message in your model. For longer prompts, paste into the system prompt field if your interface has one (Claude desktop, ChatGPT custom GPTs, API playground).
  6. Iterate. The prompts are starting points. If the output is close but not right, modify the constraints (more specific output format, tighter word limit, different tone) rather than re-prompting in chat.
  7. Found a prompt that works really well after your edits? Save it locally — these will keep changing as model behavior changes, so your version is the reliable one.
Why "by role" instead of "by task"?

Most prompt libraries organize by task — "writing", "coding", "summarization". That's how generic prompts get written: pull a verb, attach an adjective, ship it. But the way professionals actually use AI is by job context: "I'm a PM, I'm trying to ship a feature, what's the cheapest way to validate this hypothesis?" The answer depends on the role, not just the task.

Role-based grouping also makes the library skimmable. If you're a teacher, you don't want to scroll past coding prompts to find lesson-planning ones. The count next to each role (Product Manager · 10, Engineer · 10, etc.) tells you what's behind the tab before you click, so you can tell at a glance whether the library has depth in your area.

Model selection — Claude, GPT-5, or Gemini?

Each prompt carries a model badge — Claude, GPT-5, or Gemini — based on which model the author found gave better output. Claude tends to follow structured-output constraints more faithfully and is more honest about uncertainty, which is why most analytical prompts (code review, decision memos, postmortems) prefer it. GPT-5 is stronger on creative range and on instructions involving lots of formatting rules at once, which is why most writing-heavy and structured-list prompts (landing page skeletons, content briefs, press releases) prefer it. Gemini is rarely the top pick here but works well as a backup for long-context tasks.

The badges are recommendations, not requirements. All prompts work with all three. If you're paying for one model, use that one and don't worry about the badge.

FAQ

Are these prompts AI-generated or hand-written?

Hand-written. Each one was authored by someone working in that role (or with a domain expert) for a specific real task they kept running into. AI-generated prompt libraries drift into vague language because the model optimizes for plausibility over usefulness — these don't have that problem.

What's the difference between this and the Prompt Personas library?

Prompt Personas is a 1,384-prompt library sourced from the open-source awesome-chatgpt-prompts dataset, organized by topic (Development, Writing, Business, etc.). This Library by Role is much smaller (50 prompts) but every one was hand-written for a specific professional job-to-be-done. If you want maximum breadth, use Personas. If you want a curated starting point for your specific role, start here.

Can I use these in the OpenAI API, Anthropic API, or via tools like Cursor and Continue?

Yes. The prompts are plain text — paste them anywhere a model takes a system prompt or first user message. For Cursor/Continue and other coding agents, the Engineer role's prompts (especially code review, refactor plan, debug rubber-duck) are designed to drop into the system-prompt field.

How do I customize a prompt for my specific situation?

Three places to edit. (1) Replace the placeholders in double curly braces — these are required inputs (e.g. {{feature}}, {{topic}}). (2) Tighten or loosen the output format — if you want bullets instead of paragraphs, say so. (3) Adjust the constraints at the end — if the prompt says 'under 200 words' and your context needs 500, change the number. Don't be precious; the prompt is a starting point.

Why are some prompts recommended for Claude and others for GPT-5?

Different models have different strengths. Claude handles analytical work (code review, structured critique, decision memos) more faithfully — it sticks to the output format and is honest about uncertainty. GPT-5 is stronger on creative range and multi-step formatting (landing pages, sequenced emails, full content briefs). Gemini is best for long-context tasks. The badges are starting points, not rules.

Will these still work as models update?

Mostly. The prompts are structured around durable elements (role definition, clear task, explicit output format, constraints) — these don't change between model versions. What does change is whether a model strictly follows instructions or hedges. We re-test the library every quarter against the latest Claude, GPT, and Gemini and adjust the recommended-model badges accordingly.

Can I copy these for my team's internal prompt library?

Yes. The prompts on this page are free for any use, including in internal company tooling. Attribution is appreciated but not required. We'd love to hear which ones worked best for your team so we can improve the library — reach out via the contact link.

Related tools

Related tools

See all ai & prompts tools →
Looking for something else? Browse all 155 tools.