What is a system prompt?

A system prompt (also called a developer message) is the instruction set that defines the assistant's role, tone, constraints, and output format for an entire conversation. It is more durable than a single user message and shapes every reply the model gives.

Can I use these commercially?

Yes. These are generic, reusable templates with no proprietary content. Adapt the bracketed placeholders to your product, then test them on real inputs before shipping.

Why are placeholders in brackets?

Brackets like [product name] or [tone] mark the parts you must replace with your own specifics. Leaving them in tells the model the instruction is incomplete, so always fill or delete them.

How do I make a prompt more reliable?

Be explicit about output format, add edge-case handling ("if the question is out of scope, say X"), and give one or two examples. Then run the prompt through an eval set and iterate — no template is perfect on the first pass.

System Prompt Library

System prompt library

A good system prompt does most of the work: it fixes the assistant’s role, tone, boundaries, and output format before the user ever types a word. This library collects clean, reusable system prompts across the roles teams reach for most — coding helpers, support bots, tutors, analysts, writers, researchers — so you can start from a solid baseline instead of a blank box.

How it works

Every entry is a self-contained system prompt with bracketed placeholders marking the parts you customise. Filter by role, optionally narrow by output style (concise, structured, detailed), or type a keyword to search names and bodies. Click copy on any card and paste it into the system / developer message slot of your API call or chat UI. Nothing is uploaded — the whole library lives in the page.

Tips and notes

Replace every bracketed placeholder before use; an unfilled [product name] confuses the model. The biggest reliability wins come from three additions: specify the exact output format, add explicit out-of-scope handling, and include a worked example or two. Keep system prompts focused — one role per prompt — and move volatile details (today’s date, the user’s plan tier) into the user message or a tool result rather than baking them into the system prompt. After adapting a prompt, test it against a handful of real and adversarial inputs before relying on it in production.