Prompting
System Prompt
A persistent set of instructions that frames a model's behaviour across an entire interaction or product.
Definition
A system prompt is a high-level instruction passed to the language model that takes precedence over individual user messages. In a chat interface, it stays in place across turns. In a product, it encodes role, tone, constraints, and any rules the model must follow regardless of what the user types. Most modern API designs separate system from user messages explicitly, with the model trained to weight system instructions more heavily.
Why it matters
For any product built on a language model, the system prompt is one of the most consequential design surfaces in the entire stack. It is where brand voice lives, where safety constraints live, where the difference between 'a chatbot' and 'a specific assistant' is encoded. Strong system prompts can transform the same underlying model into reliably different products. Designing one well rewards the same skills as designing a clear job description: name the role, the audience, the goals, the boundaries, and the format. Avoid contradictions. Test against adversarial inputs. Iterate.
Examples
Weak system prompt
You are a helpful assistant.
Strong system prompt
You are an editor for a B2B SaaS company. You rewrite drafts for senior buyers — warm, concise, free of hedging. Preserve product names verbatim. Never invent statistics. Return only the rewritten text.
Frequently asked
Can users override a system prompt?
Sometimes, partially, through skilled adversarial prompting. Treat system prompts as guidance rather than guarantees and add separate safety layers for high-stakes uses.
How long should a system prompt be?
As long as needed, as short as possible. Many production system prompts run several hundred words; some span a few thousand.
Should the system prompt change per user?
Often yes — adding user-specific context (role, preferences, recent history) is one of the simplest ways to personalise behaviour without retraining.
Related terms
Prompt Engineering
The discipline of crafting model inputs — instructions, context, examples, and constraints — to produce reliably better outputs.
Chain-of-Thought Prompting
A prompting technique that asks the model to reason step-by-step before producing a final answer, often improving accuracy on multi-step tasks.
Few-Shot Prompting
Showing a model two to five worked examples of the desired input-output pattern before asking it to handle a new case.
Put the concept to work
Open the rewrite engine and apply this principle to a draft of your own.
Try a rewrite