OpenAI’s message is simple: treat ChatGPT like a collaborator
OpenAI is putting new emphasis on personalization as a practical way to get more relevant results from ChatGPT. In a new Academy guide, the company says the system works best when users treat it less like a search box and more like a collaborator, then give it stable context about role, preferred tone, output format and recurring needs.
The guide centers on two existing personalization tools: custom instructions and memory. Together, they represent OpenAI’s current answer to a common complaint about general-purpose AI assistants: that they can be useful in a single conversation but inconsistent across repeated work unless users restate preferences again and again.
Custom instructions set the default working style
OpenAI describes custom instructions as the place where users define what ChatGPT should know about them and how it should respond in new conversations. The examples it gives are intentionally practical rather than technical. Users might specify their role and responsibilities, ask for a concise or formal tone, request particular output formats such as bullets or copy-ready drafts, or add process guardrails such as asking clarifying questions when requirements are unclear.
The company’s framing is important. It recommends using custom instructions for stable preferences, the kind of context that does not change from one conversation to the next. That could include profession, team function, writing style or default structure. The idea is to shift recurring setup work out of individual prompts and into a standing profile.
For users, that reduces repetition. For OpenAI, it is also a way to make ChatGPT feel less generic and more dependable without requiring a specialized custom model for every use case.
