🇺🇸 United States · ChatGPT Memory — How ChatGPT Remembers You

Status: 🟩 COMPLETE 🟦 LIVING Section: 10 — AI and LLMs


Vendor	OpenAI
Country/origin	🇺🇸 United States
Recommended for AUS?	✅ Yes — useful feature with appropriate privacy controls
Privacy summary	Memories stored on OpenAI servers; you control what’s saved; can disable at any time; memories may inform model behaviour but are not used for training when opted out
Free tier	✅ Available on free ChatGPT (limited memory size)
Paid tiers	ChatGPT Plus and above have larger memory capacity
First released	Memory feature launched in beta February 2024; general availability April 2024; expanded throughout 2024–2026
Last reviewed	June 2026
Official site	https://help.openai.com/en/articles/8590148

What it is

ChatGPT Memory is a feature that gives ChatGPT the ability to remember information about you across conversations — your preferences, ongoing projects, important details about your life and work, and stylistic preferences.

Without memory, every new chat starts fresh — ChatGPT knows nothing about you. With memory enabled, ChatGPT can:

Remember you’re an Australian, that you live in Melbourne, that you’re learning Spanish
Recall that you have a teenager named Sarah and a dog called Biscuit
Know you prefer concise responses
Remember you’re working on a small business project
Apply your preferences across all future conversations without you having to re-explain

This makes ChatGPT feel more like a personal assistant who actually knows you, rather than a stranger every conversation.

How it works

There are two main ways ChatGPT remembers things:

1. Saved memories

ChatGPT proactively saves notable details from your conversations. When it does, you see a small “Memory updated” notification. Examples of what it might save:

“User is based in Sydney, Australia”
“User is a small business owner in the hospitality industry”
“User prefers responses to be concise and skip preamble”
“User is learning to code in Python and is at a beginner level”

You can see all saved memories in Settings → Personalization → Memory → Manage Memories and delete any individual memory.

2. Reference Chat History (introduced 2025)

Beyond explicit memories, ChatGPT can now reference your past conversations to provide context-aware responses — even without those memories being explicitly saved. If you mentioned a project in a chat last week, ChatGPT can refer back to that.

This is a more comprehensive form of memory: not just specific facts, but the full context of your interaction history.

How to use it (Australian users)

Turning it on

Memory is on by default for new users. To check:

ChatGPT app/web → Settings → Personalization → Memory
Toggle Memory on (or off if you prefer)
Toggle Reference saved memories and Reference chat history independently

Manually adding a memory

Tell ChatGPT directly: “Please remember that I’m vegetarian and avoid mushrooms.” ChatGPT will explicitly save that.

Reviewing memories

Settings → Personalization → Memory → Manage Memories shows everything ChatGPT has saved. Review and delete anything that’s wrong or that you don’t want stored.

Asking ChatGPT what it remembers

Try: “What do you remember about me?” ChatGPT will summarise what’s in its memory.

Forgetting things

Three ways:

Open Manage Memories → Delete specific items
Tell ChatGPT: “Please forget that I’m working on the X project.”
Settings → Personalization → Memory → Clear all memories

Temporary chats

For sensitive conversations you don’t want remembered, use Temporary Chat (top of new chat → toggle Temporary Chat). These conversations:

Aren’t saved in your history
Don’t update memory
Aren’t used for model training
Disappear after the chat ends

This is useful for: sensitive personal topics, work-confidential discussions, one-off questions you don’t want shaping future responses.

What’s actually useful about memory

After using it for a while, the genuinely useful uses become clear:

Your role and context: ChatGPT can frame technical responses for your profession (designer, accountant, doctor)
Your goals: Long-running goals like learning a language or building a business shape what ChatGPT suggests
Your style preferences: Whether you want bullet points or prose, formal or casual, concise or thorough
Your tools and stack: Knowing you use Mac, Notion, Tailwind CSS, etc., shapes recommendations
Your personal context: Family, location, interests — make conversations feel personal
Your accessibility needs: “Speak slowly in voice mode” or “use larger formatting” persistently applied

Privacy considerations

What’s saved

ChatGPT memories are stored on OpenAI’s servers, in your account. They’re tied to your specific account — not shared with other users.

Who can see your memories

You can: Always, via Manage Memories
OpenAI staff: Have access for support, abuse prevention, and trust & safety — typically not on demand but for specific issues
Other users: Cannot see your memories
Other apps via API: Memories are separate from API usage

Training use

On free ChatGPT, content may be used for training unless you opt out via Settings → Data Controls → “Improve the model for everyone” → off.

On ChatGPT Plus / Pro / Enterprise, content is generally not used for training (verify current terms).

Memory specifically is not used for training per OpenAI’s documentation.

What you should NOT save to memory

Treat memories as written-down notes about yourself stored on a third party’s server. Don’t memorise:

Passwords or credentials
Financial account numbers
Personal identifying information you wouldn’t be comfortable in a data breach
Sensitive health information you don’t want associated with your AI account
Client/customer personal details (Australian Privacy Act issues)

Australian Privacy Act implications

If you’re using ChatGPT in a business context and memory is enabled, you may be:

Storing personal information about clients on OpenAI’s servers
Potentially involving APP 8 (cross-border disclosure) obligations
Creating a record of decisions or interactions that may be subject to records management requirements

For business use, consider whether memory should be enabled and what information is being captured.

Gotchas

Memory can be incorrect. ChatGPT may infer things about you that are wrong. Review Manage Memories periodically and delete incorrect items.
Memory ≠ persistent context window. Memory captures specific facts, not the full content of every conversation. ChatGPT doesn’t remember every word you’ve ever exchanged — it captures specific summary points.
Free tier has limited memory. The free version has a smaller memory capacity than Plus. Heavy users may bump into limits.
Memory carries across all conversations. Saved memories influence every chat, including ones you might not want them to. If you’re using ChatGPT for personal AND professional work, this can blur the lines.
Reference Chat History is broader than memory. Even after deleting memories, ChatGPT may still reference your conversation history. To fully remove influence, you need to delete chats AND disable Reference Chat History.
Memory doesn’t transfer between accounts. If you use ChatGPT on multiple accounts (personal, work), memories are separate. Plus, memories don’t transfer to Claude or Gemini.
Shared devices. If multiple people use one ChatGPT account, the memory captures details about all of them — and the AI may get confused about who’s currently chatting.

Sources

OpenAI ChatGPT Memory help documentation
Memory feature announcement (February 2024)
Reference Chat History announcement (2025)
OpenAI Privacy Policy: openai.com/policies/privacy-policy
OpenAI Data Controls documentation

Tech & AI, Explained

Explorer

chatgpt-memory