If you opened the ChatGPT app this week and noticed it acting a little more like it actually knows you, that's not your imagination.
OpenAI rolled out an update that overhauls the memory system inside its mobile app, and the change is bigger than the patch notes suggest.
The headline feature is something the company calls expanded memory.
Previously, ChatGPT could remember bits and pieces you told it — your name, your job, maybe that you hate cilantro.
The new version stitches those fragments into something closer to a running profile, pulling context from past conversations and applying it automatically to new ones.
Ask for dinner ideas and it may already know you're cooking for two.
Ask for a workout plan and it might skip the exercises you complained about last month.
In practice, it's a shift in how the app treats your data.
Every chat you've ever had becomes raw material for a model that gets a little more tailored with each exchange.
The app now surfaces a memory panel where you can review, edit, or delete what it has stored about you.
Here's where the pattern gets interesting.
The same update tightened integration with voice mode and pushed more of the experience toward a persistent assistant that lives across sessions rather than a blank chat box you reset each time.
Every major AI company is racing toward the same destination: an assistant that feels less like a tool and more like a presence.
Memory is the glue that makes that possible.
Google has been pushing Gemini deeper into Android, Apple is weaving its own intelligence layer into iOS, and Samsung is bundling AI features directly into its phones.
Standalone apps like ChatGPT are fighting to stay the default destination rather than getting absorbed into the operating system itself.
Memory is one of the few features a third-party app can offer that a built-in assistant struggles to match — at least for now.
For American users, the practical question is simpler than the philosophical one.
Does the update make the app more useful?
The context carryover genuinely reduces repetitive prompting, and the memory controls are more transparent than what most competitors offer.
But there's a tradeoff nobody at OpenAI is eager to advertise: the more the app remembers, the more there is to lose if that data is ever exposed, subpoenaed, or repurposed.
Open the settings, find the memory section, and actually read what's stored.
Delete anything you wouldn't want attached to your account indefinitely.
Turn off memory for sensitive conversations before you start them, not after.
And remember that "delete" in one part of the app doesn't always mean gone everywhere.
The broader story is that we're watching the slow conversion of a chatbot into an infrastructure.
ChatGPT started as a novelty, became a utility, and is now positioning itself as something closer to a personal operating layer.
You just don't usually see them being laid.
My take: memory is the feature that will decide whether AI assistants feel magical or creepy, and the line between those two is thinner than any company wants to admit.
OpenAI is betting you'll trade a little privacy for a lot of convenience.
Final Thoughts
That bet has paid off before, but it's worth asking how much of yourself you're willing to hand over for a smarter autocomplete.