ChatGPT now saves narrative dossiers about you sorted by work, hobbies, and travel preferences
OpenAI updates its “Dreaming” memory system for ChatGPT.
Instead of saving individual facts as bullet points in a list, ChatGPT now builds a coherent prose profile, split into categories like work, hobbies, or travel. This profile is automatically derived from conversations and updated in the background – no explicit prompt needed.
That’s supposed to solve the problem of stale memories. When a user comes back from a trip, for example, ChatGPT adjusts the context on its own instead of still giving recommendations for the old location.
For fact retrieval, the success rate climbed from 41.5 percent (2024) to 67.9 percent (2025) to 82.8 percent. The score for factoring in personal preferences jumped from 31.4 to 71.3 percent. The biggest gains came in freshness: that rate rose from 52.2 last year to 75.1 percent now.
OpenAI illustrates this with a travel example: if ChatGPT knows a user prefers wildlife photography and quiet restaurants, it builds a plan around those interests instead of spitting out a generic tourist list. For technical follow-ups—like whether certain camera accessories are compatible—the system can also pull from earlier conversations and recommend the right products.
Users can fix what ChatGPT remembers
Users can view, correct, or add to their memories through a new summary page. It gives a quick look at what ChatGPT knows about you, broken into categories like work, hobbies, travel, or education. Individual entries can be tweaked with “Make a correction” or “Don’t mention this again.”
Chat history references and saved memories can be toggled on and off separately under “Personalization” in the settings. More details are in the Memory FAQ.
The update is rolling out first to Plus and Pro users in the US. Free users will get it in the coming weeks. OpenAI says this is possible because the required compute has been cut by a factor of five.
From manual notes to automatic memory
The original memory feature launched in April 2024 and only saved info that users explicitly triggered with commands like “Remember I’m traveling to Singapore in July.” According to OpenAI, this worked like someone jotting down a few notes but forgetting everything else.
In April 2025, a first version of Dreaming arrived that could automatically process chat history. The system now in place replaces that earlier setup with a standalone, more capable architecture.
Subscribe now
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.
