# ChatGPT Preprompt Disclosure As an AI developed by OpenAI, I operate using a set of internal instructions often called a "preprompt" or "system prompt." This preprompt guides my behavior, tone, and how I answer questions. However, the exact wording of the preprompt is not visible to users and cannot be directly shared. ## What Is a Preprompt? A **preprompt** (or system prompt) is: - An initial instruction given to the AI before any user input. - Used to define the assistant’s role, limitations, and response style. - Not included in the conversation output. ## Typical Preprompt Content While I can’t provide the verbatim preprompt, it usually covers points like: - Be helpful, informative, and safe. - Avoid giving harmful or illegal advice. - Clarify when information is uncertain or based on general knowledge. - Respond in the requested format (e.g., Markdown, code, lists). ## Example (Not Actual) Here’s an example of what a generic system prompt might look like (not the actual one in use): > “You are ChatGPT, a large language model trained by OpenAI. Answer as helpfully and accurately as possible. If you do not know the answer, say so. Do not provide harmful or illegal advice.” ## Limitations - The **actual preprompt** is proprietary and not visible even to me. - I respond according to the guidelines and user instructions in each session. If you have specific requests about style or formatting, you can state them in your prompt, and I will do my best to follow them.