System Prompt

A system prompt is the "instruction manual" an AI follows by default in every conversation. You never see it, but before you type your first message, the AI has already read a long set of instructions — things like "You are an assistant for X. Respond politely. Refuse dangerous requests. Today's date is Month X."

Think of it like a script's character notes for a play. The same actor (the language model) becomes a completely different character depending on what's written in the opening character brief (the system prompt). The differences in tone and personality between ChatGPT and Claude come not just from the underlying models, but from differences in these instruction sheets.

It shows up in articles in two main ways. One is leak/exposure controversies — when a hidden instruction sheet gets exposed, it becomes news because it reveals "what this AI was told not to do." The other is security — attacks that use clever inputs to make an AI ignore its instructions (prompt injection) are a recurring topic in AI security coverage.

"The list of prohibited topics revealed in the leaked system prompt" — often covered as a document showing how AI companies control their models.

1. Find the "Custom Instructions" section in your chatbot's settings.
2. Write something like: "Always answer in three sentences or fewer. If you use jargon, immediately explain it in plain language."
3. Ask anything in a new conversation — you'll notice every answer's tone has changed. What you just wrote is your own personal system prompt — your account-level version of what companies set for the entire model.

The overlapping names make them sound like two words for the same thing. In reality, what differs is who writes it and when. A prompt is what you type right now; a system prompt is the instruction the AI has already read before you type your first letter. Part of why ChatGPT and Claude sound different isn't just the underlying model — it's this instruction layer.