What is an AI Prompt?

What a Prompt / system prompt really is

The word Prompt literally means a “command” or “the text you feed the AI.” But in the context of setting a role, we usually mean the system prompt (sometimes called a persona or custom instructions) — a special instruction set “before” the conversation that defines who the AI is, what style it answers in, and what rules it must follow.

Unlike the messages you type one at a time (user messages), a system prompt is set once and applies for the entire conversation. It's like briefing a team before the work starts: “Today you're the editor — polish the language, keep it professional.” Every answer after that stays within that frame.

How a system prompt works

When you chat with an AI, the messages are organized into “layers” with different weight:

• System prompt — the instruction setting the role/style/rules, which the AI treats as the main frame throughout the conversation.
• User message — the question or command you type in each time.

Roughly how it goes in practice:

1. You set a system prompt (e.g. “You are a math tutor who explains step by step”)
2. Every time you ask, the AI reads the system prompt first, then answers in that role and style
3. As long as you don't start a new conversation, the frame stays — so you don't have to repeat the role in every message
4. Open a new chat (or change the system prompt) and the role resets or changes to the new one

What a Prompt lets AI do

A well-set Prompt steers the direction of the AI's answers instantly, without changing the model or writing code — for example:

• Take on a specialized role — a Thai editor, a tutor, a first-line legal advisor, a copywriter
• Lock in style and tone — short and concise, formal, playful, or beginner-level vocabulary
• Define the output format — answer as a table, bullets, JSON, or 1-2-3 steps
• Set rules / prohibitions — e.g. “don't guess; if you don't know, say so” or “answer in Thai only”
• Provide starting context — company info, target audience, or specific terms it must use

A Prompt sets the AI's role for the conversation, while AI Rules set permanent project-wide standards and AI Agents are specialized helpers that run on their own.

Real-world prompt examples

System prompts people use often, by task:

• Thai editor — proofreads and edits language, tightens it, makes it flow, while keeping the original meaning
• Step-by-step tutor — explains gradually, gives examples, then asks questions to make the learner think along
• Meeting-notes assistant — summarizes into topics + a list of action items with owners
• Copywriter — writes 3 short captions in different tones for a defined target audience

A full example of a short, ready-to-use system prompt:

You are a professional Thai editor
Task: proofread and edit the user's text to be correct and readable
Style: concise, polite, no filler; always keep the original meaning
Output: (1) the edited text (2) a bullet summary of what changed
Rules: don't add new content the user didn't write; if unsure, ask first

See curated prompt sets, with badges for which task they fit, in the Prompt library.

Where can you set a prompt (Claude, ChatGPT, Cursor)

Almost every major AI tool lets you set a system prompt / persona — they just call it different names:

• Claude — set it in the system prompt field (via the API or Claude Projects), and Claude likes prompts structured with XML tags like <role>, <task>
• ChatGPT — use “Custom Instructions” or build a Custom GPT to embed the persona in that chat
• Cursor / Codex and coding tools — set “rules for AI” or a system prompt so the coding assistant works in the style you want

How to start using Prompts

Setting a system prompt for the first time takes 3 steps:

1. Pick a prompt that fits the job — want AI to help proofread? Choose the editor one. Open the Prompt library and browse by task.
2. Paste it into the system / custom instructions field — copy the prompt into Claude/ChatGPT/Cursor's settings, following each tool's guide.
3. Try it and refine — test a real task; if the answer isn't quite right, sharpen the prompt, e.g. add an example or a constraint.

Don't want to write your own? Pick a Loadout with a Prompt — bundled by task, ready to use.

How to write prompts that work

An effective prompt isn't about writing a lot — it's about being clear and complete. A formula that works for almost any task is Role + Task + Context + Format + Constraint:

• Role — say who the AI is, e.g. “You are a Thai editor”
• Task — say clearly what to do, “edit the text so it flows”
• Context — who reads it, what the goal is, what info it needs to know
• Format — what the result should look like: table/bullets/steps
• Constraint — rules/prohibitions like length, language, or “if you don't know, say so”

Extra techniques that improve accuracy: add examples (few-shot) so the AI sees what a good answer looks like, and for Claude, structuring with XML tags helps the model separate the parts clearly:

<role>You are a Thai news-summary assistant</role>
<task>Summarize the given article into 3 short bullets</task>
<constraint>
- Use plain, easy-to-understand language
- No personal opinions
- If the article is unclear, state "not enough information"
</constraint>
<example>
Article: ... → Summary: • ... • ... • ...
</example>

Prompt vs Rule vs Agent

These three are often confused, but they do different jobs:

Read on: What are AI Rules · What are AI Agents · What are AI Skills

Loadouts with a Prompt

In practice you rarely use a prompt alone — you combine it with MCP/Skills/Rules to finish one job. A bundle assembled like that is called a Loadout.

For example, a content-writing Loadout might include a writer system prompt, a style-polishing Skill, and a brand-tone Rule in one set — ready to use without assembling it yourself.