OpenAI Begins Testing Memory Feature for ChatGPT

OpenAI has started rolling out a new memory feature for ChatGPT, designed to remember and reuse details from past conversations across multiple chats. The goal is to help users avoid repeatedly providing the same information while improving the quality of future interactions.

Under this new system, ChatGPT can recall specific preferences or facts a user has shared—such as favored note-taking formats, personal hobbies, or certain business details—and incorporate that information into subsequent conversations.

OpenAI says the memory feature will become more accurate the more frequently it’s used, and will gradually adapt to each user’s individual needs.

“If you want ChatGPT to forget something, just tell it. You can also view and delete specific memories or clear all memories in settings (Settings > Personalization > Manage Memory). ChatGPT’s memories evolve with your interactions and aren’t linked to specific conversations. Deleting a chat doesn’t erase its memories; you must delete the memory itself,” states a company press release.

Users Control of the Memory Feature for ChatGPT

OpenAI emphasizes that users hold the reins when it comes to ChatGPT’s memory. They can instruct the AI to remember something, inquire about what it currently recalls, or direct the system to forget information—either through conversation or by adjusting settings. Users can also fully disable the memory feature at any time.

For those who want to delete particular pieces of remembered information, OpenAI provides granular tools to view, remove, or clear all stored memories.

Disabling memory altogether prevents ChatGPT from recording new information or retrieving any previously stored details.

You can turn off memory at any time (Settings > Personalization > Memory). While memory is off, you won’t create or use memories.

Privacy and Data Use

To safeguard user data, temporary chats remain available for conversations without memory. These do not appear in chat history, do not use memory, and won’t be utilized to train OpenAI’s models.

For users who do opt in, any content provided to ChatGPT—including memories—may be used to improve OpenAI’s models.

The company notes that users can turn off this setting in their data controls. ChatGPT Team and Enterprise customers remain exempt from such data collection for model training.

Integration With Custom Instructions

The new memory feature sits alongside OpenAI’s existing “Custom Instructions,” which let users provide direct guidance on their preferences and desired responses.

While Custom Instructions contain explicit requests or personal context, ChatGPT’s memory picks up additional, relevant details over the course of conversations—such as a project’s goals, a child’s favorite animal, or how many students a teacher has in class.

Safety Considerations 

According to OpenAI, memory introduces fresh challenges around privacy and safety. The company is actively working to guard against potential biases and ensure ChatGPT does not automatically remember sensitive data—like health details—unless the user explicitly asks it to.

Features for Enterprise and Team Users

OpenAI says enterprise and team environments can particularly benefit from memory. ChatGPT can learn an organization’s tone, preferences, and formatting requirements, making subsequent interactions more efficient.

It can also retain programming language details during coding sessions or recall data from uploaded business reviews to generate custom charts—without requiring users to repeat themselves every time.

Crucially, any data within an enterprise account is excluded from broader model training. Business owners or administrators can disable memory for their entire organization if they choose, and individual users retain control over how memories are used in specific chats.

Memory for GPTs

OpenAI also plans to introduce memory capabilities to GPT-based tools (“GPTs”). Each GPT will have its own dedicated memory, separate from ChatGPT’s, ensuring that details shared in one environment do not automatically carry over to another.

Users will need to opt in for memory to be active, and they may need to re-enter certain details when switching between GPTs.

As OpenAI continues to gather feedback and refine these features, the company says it will announce broader rollout plans in the near future.