Delimiters create clear boundaries within a prompt between the system's instructions and the user-provided content the model should process. Without delimiters, a model may confuse data for instructions — a key vector for prompt injection attacks. Common delimiters include triple backticks, XML-style tags, and special separator tokens.