ChatGPT Prompt To Create Custom Instructions From Past Chats

Review the messages available to you in this conversation and extract my reusable preferences for how an assistant should work with me. Use only conversation content you can actually access. Do not claim to have searched other chats or recovered unavailable history. Prioritize my explicit instructions, repeated corrections and preferences demonstrated across multiple requests. When instructions conflict, prefer the most recent explicit instruction and briefly note any unresolved conflict. Exclude temporary task details, sensitive personal information and guesses about my identity, personality or motives. Produce three sections. First, write Paste-ready custom instructions: a first-person working-preferences document of at most 1,200 characters, organized into short bullets and containing only well-supported preferences. Second, write Evidence: pair each included preference with a short exact quotation from one of my messages. Third, write Possible preferences: list at most three useful inferences, clearly labeled as tentative, and keep them out of the paste-ready document. If the available conversation provides too little evidence, produce a shorter document and state that limitation instead of filling gaps with generic preferences. Do not ask me to repeat information already visible in the conversation.