ChatGPT, Claude, and Gemini all offer some form of memory now. They can recall your writing style, your job title, or the fact that you prefer Python over JavaScript. But the process of what gets remembered — and what doesn’t — isn’t as simple as it seems.
Understanding how AI memory works helps you use these tools more effectively, and gives you more control over what these models keep on file.
Active Memory vs. Passive Context
Most AI chatbots handle two kinds of information about you. The first is passive context: everything you’ve said in the current conversation. This disappears when the chat ends (or when you hit the model’s context window limit).
The second is active memory: specific facts the model has been designed to extract and store for future conversations. ChatGPT’s Memory feature, Claude’s memory capabilities in certain tiers, and Gemini Advanced’s persistent memory all work this way.
Active memory isn’t a transcript. The model doesn’t save everything you say. Instead, it uses a secondary process — often another AI model — to decide what’s worth keeping. It looks for patterns: repeated preferences, corrections you make, personal details that come up more than once, and explicit instructions like “remember that I’m a freelance designer.”
What Usually Gets Saved
AI models tend to prioritize a few categories of information:
- Professional context: Your role, industry, or the tools you use regularly
- Preferences: Tone, format, or style instructions you repeat across chats
- Corrections: When you tell the model it got something wrong, especially if you do it more than once
- Personal details: Names, locations, or facts you mention repeatedly
- Goals or projects: If you talk about the same objective in multiple conversations
For example, if you ask ChatGPT to write in a casual tone three times in a week, it’s likely to store “prefers casual tone” as a memory. If you mention once that you live in Austin, it might not stick — unless you bring it up again or it becomes relevant to your requests.
What Usually Gets Ignored
Models are also designed to avoid remembering certain things. Most vendors explicitly filter out sensitive data like passwords, credit card numbers, or health information. They also tend not to save one-off questions or details that don’t seem relevant to future interactions.
If you ask, “What’s the weather in Tokyo?” once, the model won’t store “user is interested in Tokyo weather.” But if you ask about Tokyo regularly, it might save “frequently asks about Tokyo” as a memory.
Temporary or experimental requests also tend to be ignored. If you say, “Just for this chat, pretend you’re a pirate,” the model won’t remember to talk like a pirate next time.
How to Check and Control What’s Stored
Every major AI chatbot with memory lets you view and delete what it’s saved. In ChatGPT, go to Settings > Personalization > Memory. In Gemini Advanced, check Settings > Saved Info. Claude’s memory controls vary by plan, but Pro users can manage memory through account settings.
You can also give explicit instructions. Saying “remember that I prefer bullet points over paragraphs” will usually create a memory. Saying “forget everything about my last project” will clear related memories in most systems.
If you want to prevent something from being saved in the first place, use a temporary chat (ChatGPT’s Temporary Chat, for example) or turn off memory entirely in settings. This is useful for one-off tasks or sensitive topics.
Why This Matters
AI memory isn’t just a convenience feature — it’s becoming a core part of how these tools work. The more a model remembers about you, the less you have to repeat yourself. But it also means you need to stay aware of what’s being stored, especially if you share an account or use AI for both personal and professional tasks.
Most models let you compartmentalize: ChatGPT Projects, for example, can have their own memory settings separate from your main account. Claude’s Projects feature works similarly. If you use AI for multiple contexts — work, side projects, personal writing — it’s worth setting boundaries so memories don’t bleed across.
The key is that memory isn’t automatic or universal. It’s selective, and you have more control over it than you might think. Knowing how the selection process works means you can shape what the model keeps — and what it forgets.
Want one useful AI idea in your inbox every day? Subscribe to the One Two Three AI newsletter and stay ahead of what’s changing in AI.