Personalizing ChatGPT

OpenAI published a rundown this week on how to get ChatGPT to actually remember what you care about, instead of treating every chat like the first date. The two levers are custom instructions, where you tell it upfront who you are and how you like answers formatted, and memory, which lets the model carry details across sessions without you re-explaining your job, your project, or your allergic reaction to bullet points.

Custom instructions have been around for a while, but the framing here is more deliberate: think of it as a standing brief you hand the assistant once, covering tone, expertise level, and constraints, so it stops guessing. Memory is the newer piece, and the more consequential one. Instead of a static profile, ChatGPT can pick up on recurring facts, preferences, and context as you use it, then apply them later without prompting.

The practical upshot is fewer redundant conversations. If you've told it you're a nurse practitioner who wants clinical answers without disclaimers, or a parent asking for recipes that avoid nuts, it should hold onto that. OpenAI frames this as a step toward responses that feel less generic and more like they came from someone who's been paying attention.

There's an obvious tension buried in this, one OpenAI doesn't dwell on: personalization and privacy pull in opposite directions. The more the model remembers, the more useful it gets, and the more data sits somewhere tied to your identity. OpenAI does give users controls to view, edit, or wipe what's stored, which matters, but the default trajectory of every major AI product right now is toward more persistent memory, not less.