AI glossary · Prompts
System prompt
In one sentence
A system prompt is the set of fixed instructions an AI model receives before any conversation starts. It sets the model's role, its tone, what it knows about the business and what it isn't allowed to do.
Also known as: system message · system instructions · custom instructions · developer prompt
What is a system prompt?
When you chat with a company’s AI assistant, the model has already been given instructions before your first message. That hidden text is the system prompt. It tells the model who it is (“you’re the front desk of a language school”), how to talk (“short sentences, friendly”) and which rules to follow (“don’t quote prices for private lessons”).
Put it to work
AI agents that look after your customers
From €49/mo + VAT
Voice, WhatsApp and email with the phone line included: agent, numbers and CRM in one fee.
See plans and pricingThe difference from an ordinary prompt is that it stays put. The user’s prompt changes with every message. The system prompt runs through the whole conversation, and it’s written by whoever sets up the assistant, not by the person using it.
Related to this term
UTIO agents
- Available 24/7 in 26 languages
- Set it up yourself for €0: built from your website
- 14-day free trial, no card needed
From €49/mo + VAT
See plans and pricingHow it works
On every turn, the model receives a bundle in three layers:
- System prompt: role, tone, rules, who to hand over to.
- History: everything said so far.
- The new message from the user, sometimes with document extracts pulled in through RAG.
All of it counts in tokens and all of it has to fit in the context window. So an endless system prompt isn’t better: you pay for it on every message.
A good system prompt spells out what to do when the model doesn’t know something (“say so and offer to pass them to a person”). It’s one layer of the guardrails, but not the only one. A user may try to get around it with prompt injection, so anything that must never happen is also blocked outside the model.
Example
An English language school in Cork puts an assistant on WhatsApp to answer questions about levels and timetables. Its system prompt, in short:
You are the assistant for an English language school in Cork.
Use short, friendly sentences in plain English.
Answer questions on levels, timetables and the placement test using the school's information.
For private lesson prices, ask for a name and phone number so the coordinator can call back.
Don't discuss other topics. If you don't know something, say so and offer to pass them to a person.
Introduce yourself as a virtual assistant in your first message.The actual timetables aren’t in the prompt: the assistant looks them up in the school’s documentation. When they change in September, nobody has to touch the instructions.
System prompts at UTIO
At UTIO you don’t write the system prompt by hand. Creator builds the agent by reading your website: it knows what you do and it answers. The agent introduces itself as a virtual assistant and starts in draft mode, so you can review its replies before you give it more autonomy.
If you’d rather we set it up, we tune the agent’s script with real conversations during the first month and connect it to your calendar and CRM. It works for both the WhatsApp agent and the voice agent.
Frequently asked questions
Can customers see the system prompt?
They can't see it, but you shouldn't treat it as secret. With the right questions, a model may end up revealing part of its instructions. So never put passwords, API keys or data you don't want leaked in there.
What happens if the user asks for something that contradicts the system prompt?
Models are trained to give system instructions priority, but that isn't an absolute guarantee. Sensitive actions should also be protected outside the model, with permissions and confirmations.
How long should a system prompt be?
Long enough to cover role, tone, rules and common cases. If it keeps growing, it costs tokens on every message and the rules start to contradict each other. Business information that changes, such as prices or opening hours, is better kept in a knowledge base.
Related terms
- Prompt
A prompt is the instruction or text you give an AI model to get something done: a question, a command, a document to summarise or a mix of all three. Much of what the model gives back depends on the prompt.
- Prompt engineering
Prompt engineering is the work of designing, testing and refining the instructions given to an AI model so that it answers well consistently, not just once.
- Guardrails
Guardrails are the safeguards placed around an AI model to keep it within agreed limits: checks on what goes in and comes out, limits on topics and actions, and rules for handing a case to a person when needed.
- Prompt injection
Prompt injection is an attack in which someone slips instructions into text an AI model reads, such as a message, an email or a web page, to make it do something other than what whoever set it up intended.
- AI agent
An AI agent is a system that uses a language model to understand a request, decide which steps to take and carry them out with real tools, such as your calendar or CRM, until the task is done rather than just replying.
Updated on