System prompt

A system prompt is the instruction passed to a large language model before any user input, used to define the model’s role, the rules it should follow, and the brand context it should generate within.

In a typical LLM call, the system prompt comes first and persists across the conversation; user messages come second and change turn-by-turn. AI marketing tools use system prompts to set the generator’s identity ("You are a copywriter for X brand"), enforce constraints ("Never use the word ‘leverage’"), and inject context ("Here are 5 past posts from this brand").

Well-engineered system prompts can run to thousands of tokens. They typically include brand voice description, signature-phrase examples, banned cliches, channel-specific format rules, and structural guidance for the desired output. The quality of the system prompt is one of the strongest predictors of output quality across AI writing tools — though it is invisible to the user.

When two AI tools using the same underlying model produce different output quality, the difference is almost always the system prompt. Tools that hide their system prompts behind feature names are often hiding short, generic prompts.