AI

System Prompts

A practical guide to understanding what system prompts are and how they shape AI behaviour in products.

What system prompts do, how they define an AI's role and constraints, and what product and design teams need to know when working with them.

22 May 20264 min read

What it is

A is a set of instructions given to an before a user interaction begins. It defines how the model should behave throughout the conversation.

typically specify the AI's role, tone, , goals, and any rules it should follow. They are invisible to the user but shape every response the AI produces.

For example, a might instruct the AI to act as a customer support agent for a specific brand, respond only in formal language, never discuss competitors, and always recommend contacting a human for billing queries.

Unlike a user's message, which changes with every , the is consistent and persistent across all conversations.

design is a core product and design responsibility — not just a technical one.

When to use it

Understand when design matters most.

It is most relevant when:

You are building a product or feature powered by an AI model
Consistent behaviour, tone, or boundaries are required
The AI needs to stay within a specific domain or scope
User safety, brand voice, or compliance require guardrails
You want to personalise the AI's behaviour for a specific context

It is less relevant when:

You are using a general-purpose AI tool with no custom integration
The AI's default behaviour is already well-suited to the task

Key takeaway

If you are building an AI feature, the system prompt is your first and most important design decision. Get it wrong and no amount of user-facing design will fix the behaviour.

How it works

Understand the basic mechanism. When a conversation begins, the is passed to the as context before any user input. The model uses this to establish a framework for how it will respond.

A well-written makes the 's behaviour more predictable, consistent, and appropriate to the context. A poorly written one leads to inconsistent responses, off-brand behaviour, or outputs that do not meet user needs.

can include role definitions, behavioural rules, tone guidance, content restrictions, output format requirements, and information the AI needs to know about the it is operating in.

What this means for designers and product teams. design is iterative. You write, test, observe the outputs, and refine — the same way you would iterate on any other product design decision.

Changes to a can have significant and unpredictable effects on the AI's . Testing after every change is essential.

also need to be maintained. As the product evolves, the changes, or new edge cases emerge, the system prompt will need updating.

What to look for

Focus on:

Role clarity — whether the AI's purpose and constraints are clearly defined
Tone consistency — whether the AI maintains the right voice across different inputs
Edge case behaviour — how the AI responds to inputs the system prompt did not anticipate
Scope enforcement — whether the AI stays within its intended domain
Failure modes — what happens when the AI receives input that conflicts with its instructions

Where it goes wrong

Most issues come from: Vague instructions produce unpredictable — precision in design is not optional.

Instructions that are too vague or open to interpretation
Conflicting rules that cause the AI to behave inconsistently
No testing across a representative range of user inputs
System prompt written once and never revisited
Treating system prompt design as a technical task rather than a design one

What you get from it

Understanding gives you:

Direct control over how an AI feature behaves in your product
A clearer brief for working with engineers on AI integrations
A foundation for consistent, on-brand AI interactions
A mechanism for enforcing constraints and reducing risk

Key takeaway

The system prompt is the instruction set for your AI's behaviour. It deserves the same rigour as any other design decision.

FAQ

Common questions

A few practical answers to the questions that usually come up around this method.

What is a system prompt?

A is a set of instructions given to an before the user starts interacting with it. It defines the AI's role, tone, constraints, and rules. Users typically do not see it, but it shapes every response the AI gives.

How is a system prompt different from a user prompt?

A user is what the person types during a conversation. A prompt is what the product team writes in advance to configure the AI's . The system prompt is persistent and consistent — the user prompt changes with every interaction.

Can users see or change the system prompt?

Not in most products. are set by the product team and are not visible to users. In some tools and API contexts they may be more visible, but in consumer products they are typically hidden.

How long should a system prompt be?

Long enough to cover the AI's role, key , and any important — but not so long that it becomes contradictory or dilutes the most important instructions. and precision matter more than length. A focused, well-written system prompt will outperform a long, vague one.

Who is responsible for writing system prompts?

It is a shared responsibility between product, design, and engineering. Product and design should own the behavioural and experience decisions — what the AI should and should not do, what tone it should use, what it should operate within. Engineering typically handles the . In practice, all three need to collaborate on testing and .

Quick take

System prompts are where you define how an AI behaves before the user ever types a word — and they are one of the most important design decisions in any AI product.

Related Services

LET'S WORK TOGETHER

Ready to improve your product?

UX, research and product leadership for teams tackling complex digital services. The work usually starts where things have become harder than they need to be: unclear journeys, inconsistent products, competing priorities, or teams trying to move forward without a clear direction. I help simplify the problem, shape the right next step, and turn complexity into something people can actually use.

Previous feedback

Will Parkhouse

Senior Content Designer

01/20