ONLINEAGENT_OPS 2026.Q3 HOME HUBSMODELSBUILDINGCOMPANIESFILMMAKINGSECURITYLIVING WITH ITSKILLSALL HUBS, ONE TERMINAL → ARTICLES CRAFT RECORD BLOG TOOLS SKILLS SEARCH /
HOME›BLOG›PROMPT VS SYSTEM PROMPT: WHAT IS THE DIFFERENCE?
BLOG · COMPARISONBEGINNER

Prompt vs System Prompt: What Is the Difference?

A prompt is the message you type; a system prompt is the builder’s hidden standing instructions that come first. Who writes each, which wins, and why.

READ8 min
WORDS1,544
SECTIONS7
TYPECOMPARISON
CHECKED19 OCT 26

A prompt is the message you type to an AI. A system prompt is the set of standing instructions the app’s builder gives the AI before you type anything, setting its role, tone and rules for the whole chat. You write the prompt and can see it; the system prompt is usually written by someone else and hidden from you.

TL;DR — THE SHORT VERSION
  • Prompt: your request, in your words, changing with every message.
  • System prompt: the builder’s instructions, placed in front of your message and kept for the whole chat.
  • Which wins: the vendors design their models to give the builder’s instructions priority over yours, but a system prompt is a guide, not a lock.
◈ IN PLAIN TERMS

Think of a shop. The system prompt is the staff handbook the manager hands over before opening: “You work in the garden department. Be friendly. Never promise refunds.” Your prompt is what you, the customer, ask at the counter. The assistant answers your question, but always within the handbook you never get to read.

Prompt and system prompt side by side
QuestionPromptSystem prompt
Who writes it?Written by: you, the person using the app.Written by: whoever built the app, or a developer using the model.
When is it read?Read: each time you send a message.Read: first, in front of your messages, at the start of the chat.
Can you see it?Visible: yes, it is what you typed.Visible: usually not, though some companies publish theirs.
What is it for?For: this task: “summarise this”, “draft that”.For: the role, rules, tone and format for every task.
Other namesAlso called: user message, input.Also called: system instructions (Google), developer message or instructions (OpenAI).

What is a prompt?

The plain point: a prompt is whatever you ask the AI, in ordinary words.

Google’s guide for its Gemini models calls prompts “natural language requests”: requests written the way people normally talk.1 It can be one line or a page of background, examples and limits. The full explanation is on prompt meaning, and what makes a good prompt covers the parts that help.

What is a system prompt?

The plain point: it is the instructions that come before yours, written by the builder, and they shape every reply.

NIST, the US government’s standards agency, describes instructions “often prepended to the model’s other input”, such as “You are a helpful financial assistant who responds gracefully and concisely....”, which “is known as a SYSTEM PROMPT.”2 “Prepended” just means placed in front.

The vendors say what belongs there. Google: “Place essential behavioral constraints, role definitions (persona), and output format requirements in the System Instruction”.1 Anthropic, which makes Claude, says a role there changes behaviour: “Even a single sentence makes a difference”.3 More on system prompt meaning.

Chat apps have one too. Anthropic says the Claude website and apps use a system prompt to give Claude up-to-date information, such as the current date, and to encourage habits like putting code in Markdown.4

What do the two look like side by side?

The plain point: the system prompt sets up the job once; each prompt asks for one piece of it.

EXAMPLE SYSTEM PROMPT — WRITTEN ONCE BY THE BUILDER
You are a patient homework helper for children aged 9 to 11. Use short, everyday words. Never give the final answer straight away: ask one question that helps the child take the next step. If a question is not about schoolwork, say kindly that you can only help with homework.
EXAMPLE PROMPT — TYPED BY THE CHILD, CHANGES EVERY TIME
What is 7 times 8?

With that system prompt, the same model that would simply give the answer should instead ask a helping question. In this site’s reading, that is why two apps built on the same model can feel like different products: the hidden instructions differ, not the AI.

OpenAI puts the split in programming terms: “You could think about developer and user messages like a function and its arguments in a programming language.” The developer message holds “the system’s rules and business logic”.5

TAKEAWAY

Put what never changes (who the AI is, its rules, its format) in the system prompt. Put what changes (this task, this file) in the prompt.

Which one wins when they disagree?

The plain point: the builder’s instructions are meant to win. But a determined attack can still get round them.

OpenAI says developer messages are “prioritized ahead of user messages”, and user messages are “prioritized behind developer messages.”6 So when a chatbot refuses a topic or sticks to one company’s products, it is often following its system prompt, not giving its own opinion.

That priority is a design aim, not a guarantee. NIST describes prompt injection as an attack that “overrides these instructions”.7 See prompt injection meaning and jailbreak meaning. If you build with AI, never put a secret in a system prompt and assume it is safe there; that is this site’s advice, following from NIST’s point.

Can you write your own system prompt?

The plain point: developers can, directly. In chat apps, the nearest thing is a settings box for standing instructions.

  • Through the API. Developers send it in its own field. In Anthropic’s own example, the system field reads “You are a helpful coding assistant specializing in Python.”8 Anthropic’s published app system prompts do not apply there.4
  • In a chat app. Custom instructions are standing preferences you write once. Claude’s help centre says they “apply broadly to all your conversations”.9 In this site’s reading, they sit alongside the app’s own system prompt; they do not replace it.
  • For an agent. If you set up an AI that acts for you, its standing instructions are the most important text you write. The agent brief gives a structure.

Both count against the same limit. Anthropic lists the system prompt and every message as taking up room in the context window.10 A very long system prompt leaves less space for the chat.

Questions people ask

Can I see a chatbot’s system prompt?

Usually not. Some companies publish theirs: Anthropic posts the system prompts for its Claude apps on its release notes page (source above).

Does the system prompt outrank what I type?

It is designed to. OpenAI says developer messages are prioritized ahead of user messages (source above). That is why some requests get a polite refusal in one app and an answer in another.

Can a system prompt stop jailbreaks?

Not reliably. NIST describes prompt injection as an attack that overrides the system prompt’s instructions (source above), so builders add other checks too. Prompt injection explains the defences.

Should I start my own prompt with “You are an expert”?

Anthropic recommends roles in the system prompt (source above). In a normal chat, in this site’s use, saying who the answer is for and what you need does more than a title; prompting myths checks the popular phrases.

Where to go next

QUICK CHECK · 3 QUESTIONS

Pick an answer to see if you got it. Nothing is recorded.

1. Who usually writes the system prompt in a chat app?

  • The person using the app, in the chat box
  • Whoever built the app
  • The AI writes it for itself after each answer
Show the answer

Answer: Whoever built the app. The page says the system prompt is the builder’s standing instructions, given to the AI before you type anything.

2. According to the page, what belongs in a system prompt rather than in each prompt?

  • The role, rules, tone and format that apply to every task
  • The one file you want summarised right now
  • Nothing; system prompts are only for safety warnings
Show the answer

Answer: The role, rules, tone and format that apply to every task. The page advises putting what never changes in the system prompt and what changes, such as this task, in the prompt.

3. Can a system prompt reliably stop jailbreaks and prompt injection?

  • Yes, the builder’s instructions can never be overridden
  • Yes, as long as it is long enough
  • No, it is a guide, not a lock, and attacks can override it
Show the answer

Answer: No, it is a guide, not a lock, and attacks can override it. The page quotes NIST describing prompt injection as an attack that overrides the system prompt’s instructions.

SOURCES
  1. 1Google AI for Developers, Prompt design strategies, last updated 13 Oct 2026, read at source 19 Oct 2026“Prompt design is the process of creating prompts, or natural language requests, that elicit accurate, high quality responses”. ↑a“Place essential behavioral constraints, role definitions (persona), and output format requirements in the System Instruction”. ↑b
  2. 2NIST, AI 100-2e2025, Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations, March 2025, read at source 1 Oct 2026 (as cited on this site’s system prompt meaning): the developer or user provides instructions “often prepended to the model’s other input”, a description of the use case such as “You are a helpful financial assistant who responds gracefully and concisely....”, which “is known as a SYSTEM PROMPT.”↑
  3. 3Anthropic, Prompting best practices, undated documentation, read at source 19 Oct 2026“Setting a role in the system prompt focuses Claude’s behavior and tone for your use case. Even a single sentence makes a difference”. ↑
  4. 4Anthropic, System prompts (release notes), undated documentation, read at source 19 Oct 2026“Claude’s web interface (claude.ai) and mobile apps use a system prompt to provide up-to-date information, such as the current date, to Claude at the start of every conversation. The system prompt also encourages certain behaviors, such as always providing code snippets in Markdown.” ↑a“These system prompt updates do not apply to the Claude API.” ↑b
  5. 5OpenAI, Prompt engineering, undated guide, read at source 19 Oct 2026“You could think about developer and user messages like a function and its arguments in a programming language.” “developer messages provide the system’s rules and business logic, like a function definition.” ↑
  6. 6OpenAI, Text generation, undated guide, read at source 19 Oct 2026“developer messages are instructions provided by the application developer, prioritized ahead of user messages.” “user messages are instructions provided by an end user, prioritized behind developer messages.” ↑
  7. 7NIST, AI 100-2e2025, March 2025, read at source 1 Oct 2026 (as cited on this site’s system prompt meaning): “A PROMPT INJECTION overrides these instructions, exploiting the concatenation of untrusted user output to the system prompt to induce unintended behavior.”↑
  8. 8Anthropic, Prompting best practices, undated documentation, read at source 19 Oct 2026, code example"system": "You are a helpful coding assistant specializing in Python." ↑
  9. 9Claude Help Center, What are skills?, read at source 13 Oct 2026 (as cited on this site’s skills vs MCP vs plugins): custom instructions “apply broadly to all your conversations”.↑
  10. 10Anthropic, Context windows, undated documentation, read at source 19 Oct 2026“Everything in the request counts toward the context window: the system prompt, every message in messages (including tool results, images, and documents), and your tool definitions.” ↑

Read at source 19 Oct 2026: Google AI for Developers, Prompt design strategies (last updated 13 Oct 2026); Anthropic, Prompting best practices, System prompts and Context windows; OpenAI, Prompt engineering and Text generation.

Reused from this site’s pages, with the read dates recorded there: NIST, AI 100-2e2025 (1 Oct 2026); Claude Help Center, What are skills? (13 Oct 2026).

The shop handbook comparison, the homework-helper example prompts and the advice marked as this site’s reading or use are this site’s own, not claims from these sources.

Free printable AI cheat sheets, and AI news every week.

Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.

SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK
aipromptgeneer://$ about this siteAI PROMPT GENEERAI explained in plain English. Every claim links to its source, with the date it was checked.NEWSLETTERWhat changed in AI each week, in plain English.SUBSCRIBE FREE ↗AI PROMPT GENEER · MACHINE ASSISTANCE DISCLOSEDEVERY ARTICLE CARRIES ITS OWN CHECKED DATE