System Prompt Library
Curated, tested system prompts for common roles — code reviewer, research assistant, copy editor, debate partner, brainstorm facilitator. Browse, copy, adapt.
Why a Curated Library, Not a Generated One
The web has thousands of 'prompt libraries' that are mostly LLM-generated suggestions, ranked by upvotes from people who never tested them. The result is a long tail of plausible-sounding prompts that fall apart in actual use. This library is small on purpose. Each entry is a system prompt that has been used in real workflows, refined over multiple iterations, and survives across the major model families. They are not meant to be copy-pasted verbatim into production — they are starting points to adapt. The value is in seeing what a working version looks like before you write your own. Use the AI Prompt Template Builder to parameterize the ones you adopt.
How the Library Is Organized
The library groups prompts by role. Each role has between one and three variants — a default version, plus tighter or looser alternates where the trade-offs matter. Roles included at launch: code reviewer (two variants — strict vs collaborative), research assistant, copy editor (with two registers — professional and conversational), brainstorm facilitator, debate partner, technical interviewer, and explainer-of-difficult-topics. Click a role to expand it: the system prompt itself, a one-paragraph explanation of what each part of the prompt is doing, and a list of when to choose this variant over the alternates. Copy button at the bottom of each entry. No accounts, no sign-in. The library content updates as new useful patterns are tested; you can spot the latest additions by the 'New' tag.
See also: when a polished system prompt produces a draft you want to keep refining locally, the AI Paraphraser is a Flan-T5-based rewriter, and the AI Grammar Checker applies sentence-level corrections.
Frequently Asked Questions
Why are there so few prompts in the library?+
Because depth matters more than count for system prompts. A library of 500 mediocre prompts is worse than a library of 20 that work. New entries are added only after testing across multiple model families and use cases.
Will these prompts work on Claude, GPT, and Gemini equally?+
The prompts are tested across all three. They behave most consistently on Claude and GPT-4-class models; on smaller or older models, you may need to simplify the role description or shorten the prompt.
Should I use these prompts verbatim?+
Better to adapt them. The library shows what a working version looks like — your context, audience, and tone will shift the wording. Treat each entry as a working template, not a final answer.
Can I submit a prompt I have refined to add to the library?+
Feedback link at the bottom of the page goes to the maintainer. The bar for inclusion is real-world use plus survival across model families. Submissions with usage notes are easiest to review.
What is the difference between a system prompt and a user prompt?+
System prompts set durable context that applies for the whole conversation — the model's role, constraints, and tone. User prompts are the per-turn requests. This library covers system prompts; user-message templates belong in the Prompt Template Builder.
Does the role affect output style as much as the task instructions?+
More than most users expect. A 'strict code reviewer' system prompt produces different feedback on the exact same code than a 'collaborative reviewer' system prompt. The role primes vocabulary, attention, and willingness to push back.
Are the prompts safe for production use?+
They are starting points, not deliverables. Production deployment needs additional layers: output validation, guardrails for harmful content, prompt-injection defenses. The library is for prototyping and personal workflows.
What about jailbreak-style prompts?+
Out of scope. The library is for legitimate role definition, not for circumventing model safety. Prompts that try to override model defaults are not included and not added on request.
Built by Derek Giordano · Part of Ultimate Design Tools
Privacy Policy · Terms of Service