If you've ever opened a new ChatGPT chat and immediately typed something like "I'm a marketing manager at a small agency, my tone is casual, I write for small business owners" you already know how tedious that gets.
You're starting from scratch. Again.
ChatGPT has a Memory feature that fixes this. It remembers what you tell it, learns from your past conversations, and starts applying that context automatically. Once it's set up properly, you stop briefing the AI at the start of every session and just get on with the actual work.
Here's what it does and how to make it work for you.
How ChatGPT Memory Actually Works
Memory runs in two ways.
The first is saved memories. These are things you explicitly tell ChatGPT to remember. You just type it into any chat: "Remember that I write for a B2B software company and my audience is non-technical buyers." It stores that and applies it going forward in every future conversation.
The second is chat history. Since April 2025, ChatGPT also references your past conversations automatically. It picks up on patterns: how you ask questions, what formats you prefer, what topics you work on. All without you doing anything extra. This is available to Plus and Pro users.
In January 2026, OpenAI upgraded this further. ChatGPT can now pull details from conversations going back a full year. And when it references a past chat, it shows a "Sources" link so you can click straight back to that original conversation.
Both features can be turned on or off separately in Settings > Personalization. You're fully in control.
Check What It Already Knows First
Before adding anything, do a quick audit. Open a new chat and type:
What do you remember about me?
ChatGPT will show you everything it's currently storing. Some of it might be useful. Some might be outdated or just plain wrong.
You can delete individual memories in Settings > Personalization > Manage Memories, or you can tell it directly in the chat: "Forget that I work in e-commerce." It removes it right away.
Do this before adding new memories. You want a clean, accurate baseline to build from.
What to Actually Tell It to Remember
This is where most people go wrong. They either do nothing or paste in a 500-word bio. Neither works.
The goal is short, specific facts that change how ChatGPT responds to you. Think about what you type at the start of a fresh chat most often. That's what belongs in memory.
Here are examples across common use cases:
Your role and context
Tell it:
Remember that I'm a freelance copywriter. My clients are mostly small e-commerce brands. I write product descriptions, emails, and landing pages.
Your preferred output format
Try something like:
Remember that I prefer bullet points over long paragraphs. Keep responses concise unless I ask for more detail.
Your tone and style
Here's a good one for writers and content creators:
Remember that my writing tone is conversational and direct. No corporate language. No filler phrases.
Ongoing projects
Add context when you're deep in a project:
Remember that I'm currently working on a content strategy for a SaaS company targeting HR managers at mid-sized companies.
Recurring task preferences
Works well for things you ask ChatGPT to do regularly:
When I ask for email subject lines, always give me five variations, not one.
You don't need to add all of these at once. Start with the two or three things you re-type the most. Build from there as you notice what you keep repeating.
What to Keep Out of Memory
Memory is not a private vault. Don't treat it like one.
OpenAI explicitly advises against storing sensitive information. That means no passwords, no financial details, no health data, no confidential client information. If something feels private, it doesn't belong here.
Also, don't try to paste in full templates or scripts. ChatGPT's memory is built for preferences and context, not exact verbatim content. If you need it to follow a specific structure, paste that structure directly into your chat each time you need it.
One thing that catches people off guard: deleting a chat does not delete the memories from it. If you shared something in a conversation and want it gone, you need to manually remove the saved memory in Settings > Personalization > Manage Memories. They're stored separately.
Memory vs Custom Instructions
These two features often get confused. They're different tools for different things.
Custom instructions are static rules you set once and they apply to every chat, always. Good for things like "always respond in English" or "never add a preamble before your answer."
Memory grows over time as you use ChatGPT. Think of it as a profile the AI builds about you rather than a rulebook you wrote.
The best setup uses both. Custom instructions handle your fixed, non-negotiable preferences. Memory handles the growing context ChatGPT picks up about your work, your habits, and your goals.
Neither one replaces good prompting. But together, they mean you're starting every chat from a much stronger position than most people. And that directly affects the quality of what you get back. If you're still getting generic responses from ChatGPT, this is often what's missing.
Use Temporary Chat When You Don't Want Memory
Sometimes you want a clean slate. No memory. Nothing stored.
That's what Temporary Chat is for. Start one from the ChatGPT sidebar and nothing in that conversation pulls from your memory or adds to it. Completely separate from your regular sessions.
Good for:
- Asking something sensitive you don't want lingering in your history
- Quick one-off tasks where your usual context is irrelevant
- Testing prompts without affecting your regular setup
You can use it any time without touching your main memory settings at all.
Keep Your Memory Up to Date
This is the step most people skip. It matters more than you'd think.
Memory gets stale. If you've been using ChatGPT for a few months, some of what it knows about you might no longer be accurate. Projects end. Jobs change. Preferences shift. If ChatGPT is still operating on old context, your results reflect that.
Every few weeks, run this:
What do you remember about me? Flag anything that might be outdated.
Update what needs changing. Delete what doesn't apply anymore. It takes five minutes. And it's exactly the kind of habit that separates people who get consistent, reliable results from AI from people who feel like the AI never quite gets them.
The AI only knows what you've taught it. Keep that current and you'll notice the difference.
FAQ
Is ChatGPT Memory available on the free plan? Yes. Memory has been available to free, Plus, Team, and Enterprise users since September 2024. The full chat history feature, including the ability to reference conversations going back a year, is available to Plus and Pro users.
Does memory carry over between devices? Yes. Memories are tied to your OpenAI account, not your device. What ChatGPT knows about you on your laptop applies on your phone too.
Can I turn memory off completely? Yes. Go to Settings > Personalization and you'll find separate toggles for saved memories and chat history. You can turn off one or both at any time.
If I delete a chat, does ChatGPT forget what I shared in it? Not automatically. Saved memories from a deleted chat stay active unless you remove them manually in Settings > Personalization > Manage Memories.
Will ChatGPT automatically remember everything I say? No. It picks up on things that seem like useful preferences or context. But you can always tell it explicitly to remember or forget specific things, and it'll act on that immediately.
Once your memory is set up and kept current, ChatGPT stops feeling like a stranger you have to brief at the start of every session. It starts feeling like something that actually knows your work. That's when the tool gets genuinely useful on a daily basis.
If you're using ChatGPT to create content and want a way to turn that output into polished presentations or documents, Gamma makes that process quick and easy. You paste your content in and it builds a clean, professional deck without any design work on your end.
