glossary

What Is a System Prompt? AI Glossary Explained

A system prompt is a hidden set of instructions given to an AI model before your conversation begins, setting its persona, tone, rules, and boundaries. Providers set a default one for consumer apps; developers and power users can often add their own on top through custom instructions or the API.

5 min read · Updated Jul 1, 2026

A system prompt is a hidden set of instructions given to an AI model before your conversation starts — it sets the model’s persona, tone, rules, and boundaries, shaping every reply without you ever seeing it. It’s the invisible layer that decides, for example, why one chatbot answers concisely by default and another explains everything in detail.

System prompt vs. user prompt

System promptUser prompt
Who sets itThe provider, app builder, or you (if customizable)You, in the message box
When it appliesSet once, before the conversation startsSent fresh with every message
What it controlsPersona, tone, rules, formatting defaults, safety boundariesThe specific question or task for that turn
VisibilityUsually hiddenAlways visible — it’s what you typed

What system prompts actually control

A system prompt typically sets:

  • Persona and tone — “You are a helpful, concise assistant” vs. “You are a formal, detail-oriented research assistant.”
  • Rules and restrictions — topics to avoid, required disclaimers, or formatting requirements (e.g., always answer in bullet points).
  • Safety guardrails — instructions around refusing harmful requests, protecting personal data, or staying within a company’s usage policy.
  • Context the model should assume — for a customer-support bot, the system prompt might include product details so the model doesn’t have to be told them every time.

Who writes the system prompt

For consumer apps like ChatGPT, Claude, and Gemini, the provider sets a default system prompt covering safety and general behavior — you never see or edit it directly. On top of that default, many apps let you add a layer of your own:

  • ChatGPT — custom instructions, applied across new conversations.
  • Claude — Projects and custom styles, which set standing context and tone for a workspace.
  • Developers building on any provider’s API — full control over the system prompt for whatever app they’re building.

Why system prompts became a bigger deal in 2026

As more people build apps and workflows on top of AI models rather than just chatting with them directly, the system prompt has become the main lever for making a general-purpose model behave like a specialized tool. A single underlying model — say, GPT-5 or Claude Opus — can power a legal-research assistant, a coding helper, and a customer-support bot, with the difference between them coming almost entirely from a carefully written system prompt rather than a different model underneath.

A simplified example

System prompt:
You are a customer support assistant for a software company.
Always be concise and polite. Never share pricing you are not
certain about. If you don't know an answer, say so and offer
to escalate to a human.

User prompt:
Can I get a refund after 40 days?

The system prompt never appears in the chat window, but it’s why the model’s answer stays polite, cautious about pricing, and willing to escalate — regardless of how the user phrases the question.

Can you see or change the system prompt?

Providers rarely publish their full, current default system prompt, since it changes as models and policies are updated. What you usually can do is add your own instructions on top through features like custom instructions or a Projects-style workspace — effectively a personal system prompt layered over the provider’s default one. See our prompt engineering guide for how to write these effectively.

Set your style once across every model with Perspective AI

Affiliate link for Perspective AI — we may earn a commission at no extra cost to you.

Common system prompt patterns

A few patterns show up repeatedly, whether the system prompt is a provider’s default or one you write yourself:

  • Role-setting — “You are a [role]” statements that establish expertise, tone, and scope before any question is asked.
  • Constraints — explicit boundaries, like refusing certain topics, avoiding specific claims, or staying within a word limit.
  • Format rules — requiring bullet points, a specific structure, or a particular language for every response.
  • Fallback behavior — instructions for what to do when the model is unsure, such as asking a clarifying question instead of guessing.

Writing your own custom instructions well usually means combining a couple of these — a role plus a format rule, for example — rather than a single vague sentence like “be helpful.”

Why this matters

Two AI apps can answer the exact same question differently mainly because of their system prompts, not because one model is “smarter” than the other. Understanding this helps explain inconsistent tone across providers — and why setting your own custom instructions, where available, is one of the highest-leverage things you can do to get consistently useful answers.

If you regularly switch between providers and want one consistent persona instead of resetting custom instructions in three different apps, an aggregator like Perspective AI lets you set your preferred style once and apply it across GPT-5.2, Claude, Gemini, and 30+ other models. There’s no free trial, but there’s no commitment required either — a Pay-As-You-Go option lets you pay only for the credits you use, with flat subscriptions starting at $14.99/mo above that.

Going deeper

Frequently asked questions

What is a system prompt, in simple terms?

It's a hidden instruction, set before your conversation starts, that tells the AI model how to behave — its persona, tone, rules, and any topics it should avoid. You don't see it, but it shapes every reply.

How is a system prompt different from what I type?

What you type is the user prompt — a specific request for that turn. The system prompt sits above it, setting standing rules that apply to the whole conversation, like 'always answer in French' or 'never give financial advice.'

Who writes the system prompt in ChatGPT, Claude, or Gemini?

The provider sets a default system prompt behind every consumer app, covering safety rules, tone, and formatting defaults. Many apps also let you add your own layer through settings like custom instructions.

Can I see the exact system prompt an AI app uses?

Not always in full. Providers sometimes publish parts of their default system prompt for transparency, but the complete, current version usually isn't public — and it changes over time as models are updated.

Can I write my own system prompt?

Yes, in most apps. Features like ChatGPT's custom instructions or Claude's Projects let you add standing instructions — such as a preferred tone or context about your work — that apply across a whole conversation or workspace.

Does a system prompt guarantee the model will follow it exactly?

Mostly, but not perfectly. Models are trained to prioritize system-level instructions, but very long conversations or unusual requests can occasionally cause a model to drift from an instruction it was given at the start.

Why does the same question sometimes get different answers from different AI apps?

Different providers use different default system prompts — one might instruct concise answers, another more detailed ones, another a specific safety posture — which is a major reason the same question can read very differently across ChatGPT, Claude, and Gemini.