What if the assistant you use every day quietly remembered that you were job hunting, that your kid has a peanut allergy, or that you were researching a medical symptom three months ago? That is exactly what ChatGPT’s memory feature does by design. It is genuinely useful when you want continuity, and genuinely unsettling when you don’t. The good news: turning it off takes less than a minute, and you can also wipe what has already been stored.
This guide walks through the exact controls, the difference between saved memories and chat history references, the mistakes people make when they think memory is off but it isn’t, and how to verify the setting actually stuck.
What ChatGPT Memory Actually Does
OpenAI introduced memory on February 13, 2024, initially to a small test group, with broader rollout to follow. The feature works in two distinct ways, and confusing them is the single biggest source of frustration.
The first is saved memories. ChatGPT stores discrete facts you mention, such as your profession, your preferred writing style, or a project you’re working on, and carries them into future conversations. The second is reference to past chat history, where the model can pull context from earlier conversations even when nothing was explicitly “saved.”
Both are controlled separately in some versions of the product, which is why someone can delete every saved memory and still find ChatGPT referencing an old chat. According to OpenAI’s own announcement on memory and new controls for ChatGPT, users were given the ability to turn memory off entirely, delete individual memories, clear all memories, and simply ask the model to forget something. (ChatGPT)
If you’re still deciding which assistant deserves your data in the first place, a comparison like ChatGPT vs Perplexity helps frame the tradeoffs between convenience and privacy across tools. (Privacy and Security)
How to Turn Off ChatGPT Memory on the Web
The web interface is the most consistent place to change this setting, and it’s where most people should start.
- Sign in to ChatGPT and click your profile icon in the bottom-left corner (top-right on some layouts).
- Open Settings.
- Go to Personalization (in some versions this is labeled Memory directly).
- Toggle off Memory and, if shown separately, toggle off Reference chat history.
- Close the panel. Changes save automatically.
Two things trip people up here. First, the menu path has shifted repeatedly across web, iOS, and Android, and it can differ by plan and region, so if “Personalization” isn’t visible, look for “Memory” as a top-level item. Second, turning off memory does not delete memories already saved. The toggle stops future use; the cleanup is a separate step covered below.
Turning it off on iPhone and Android
On mobile, tap the menu icon, then your profile, then Settings, then Personalization or Memory, and flip the toggle. The mobile apps occasionally lag behind the web version by a release cycle, so if the toggle is missing on your phone, sign in on a browser, change it there, and the setting will sync.
What happens the moment you flip the switch
Once memory is off, ChatGPT stops creating new saved memories and stops pulling from them. It also won’t use past conversations as context. You keep full access to your chat history in the sidebar; it just isn’t fed into the model automatically. The assistant also loses the personalization that made it feel tailored, which is the honest tradeoff. Expect to restate your preferences more often.
Deleting Memories That Are Already Saved
Turning off the tap doesn’t drain the bucket. Existing memories stay in your account until you remove them.
- Open Settings, then Personalization or Memory.
- Select Manage memories.
- Delete entries one at a time, or use Clear all if you want a clean slate.
- Confirm the deletion.
You can also delete memories conversationally. Telling ChatGPT to “forget that I live in Berlin” removes the specific stored item. This is fast but less reliable than the settings panel, because the model has to correctly identify which memory you mean. When precision matters, use the list.
One practical habit worth adopting: review the memory list once a month the way you’d review app permissions. People are consistently surprised by what accumulated, including offhand details they never intended to store.
Temporary Chat: The Cleaner Option for Sensitive Questions
If your goal is a single private conversation rather than a permanent settings change, Temporary Chat is the better tool. Introduced around April 2024, it creates a chat that doesn’t appear in your history and isn’t used for memory.
The catch is discoverability. In most versions it lives behind the model selector at the top of the interface or in the sidebar menu, and it’s easy to start a normal chat by accident. A useful habit is to start every sensitive conversation by explicitly opening Temporary Chat first, rather than opening a normal chat and hoping you remember to delete it later.
Temporary Chat also doesn’t sync across devices, so if you switch from laptop to phone mid-thought, the conversation won’t be waiting for you. That’s the point, but it’s worth knowing before you rely on it.
The Mistakes That Leave Memory Running
Having walked through this with plenty of users, certain errors come up again and again.
Turning off memory but leaving chat history reference on. These are separate switches in several product versions. If only one is off, ChatGPT can still draw on old conversations.
Assuming deletion equals a settings change. Deleting memories clears what’s stored. If the toggle stays on, new memories get created within a few conversations.
Changing it on one device and forgetting the rest. The setting is account-level, so it should propagate, but app versions sometimes render the toggle inconsistently. Verify on the web after changing it on mobile.
Confusing memory with model training. Turning off memory does not change whether your data is used to improve models. Those are different controls, documented separately in OpenAI’s privacy and data policies. If training opt-out matters to you, look for that setting specifically.
Expecting the assistant to announce it forgot. It won’t. You’ll only notice when it stops referencing your preferences.
Verifying That Memory Is Really Off
Don’t take the toggle’s word for it. Test it.
Start a new chat and mention a distinctive but harmless detail, such as a made-up project name. Then open a second new chat and ask whether the assistant knows anything about that project. If memory is genuinely off, it shouldn’t. If it does, something is still enabled, and you should revisit both the memory toggle and the chat history reference toggle.
You can also return to the memory management screen after a few days of use. If new entries have appeared, the setting didn’t stick, which usually means a plan-specific default or a stale app version. Updating the app and re-checking on the web resolves most cases.
For anyone weighing whether the personalization is worth the tradeoff, a broader look at how competing assistants handle context, such as this ChatGPT vs Gemini comparison, makes the differences clearer.
How Memory Compares Across Plans and Platforms
| Control | What it does | Where to find it | Notes |
|---|---|---|---|
| Memory toggle | Stops saving and using stored facts | Settings, then Personalization or Memory | Does not delete existing memories |
| Reference chat history | Stops pulling context from past chats | Same settings area, separate toggle | Often missed; check independently |
| Manage memories | Lists, deletes, or clears saved items | Settings, then Manage memories | Use “Clear all” for a full reset |
| Temporary Chat | Creates a chat excluded from history and memory | Model selector or sidebar menu | Does not sync across devices |
| Data controls | Governs model training use | Privacy settings | Separate from memory entirely |
Availability has shifted since launch. Memory reached Plus and Team users in April 2024, with Enterprise and free-tier access following over time, and the exact order and dates are worth confirming against current release notes rather than assuming. OpenAI maintains a ChatGPT release notes page that tracks these rollouts, and it’s the fastest way to check whether a feature has reached your plan.
When You Might Actually Want Memory On
It’s worth saying plainly that memory isn’t the villain here. For ongoing projects, it removes the tedium of re-explaining context every session. Writers who want consistent tone, developers working through a long build, and anyone managing a recurring task all benefit.
A middle path works well for many people: leave memory on for a dedicated work account and off for personal use, or keep it on but audit the memory list monthly. The binary framing of “on or off” misses the more practical question of which conversations deserve continuity. If you’re comparing how different assistants handle this kind of personalization, this Grok vs ChatGPT vs Gemini breakdown covers the tradeoffs in more depth.
The people who end up happiest with the feature are usually the ones who checked the memory list early, saw exactly what kind of detail gets stored, and made a deliberate call. The ones who get surprised are the ones who never looked.
Conclusion
Turning off ChatGPT memory is a two-part job: flip the toggle, then clear what’s already stored. Do both, verify with a test chat, and check back in a week to confirm nothing new appeared. Skip the second step and you’ve only half solved the problem, which is the mistake nearly everyone makes the first time.
The controls live in Settings under Personalization or Memory, and they’ve moved enough times that patience is warranted if the labels don’t match this guide exactly. For one-off sensitive questions, Temporary Chat is the cleaner tool. And remember that memory settings and data-training settings are different things, governed by different switches.
None of this requires abandoning the tool. It requires ten minutes of attention and a habit of auditing what’s stored. That’s a small price for knowing exactly what the assistant remembers about you.
Frequently Asked Questions
Does turning off memory delete what ChatGPT already knows about me?
No. The toggle stops new memories from being saved and stops existing ones from being used, but stored entries remain in your account until you remove them. Go to Settings, then Personalization or Memory, then Manage memories, and delete entries individually or use Clear all.
Is memory the same as chat history?
No, and this is the most common point of confusion. Chat history is the visible list of your past conversations. Memory is what the model carries forward and applies. You can have memory off while your chat history remains fully intact and readable.
Does turning off memory stop OpenAI from using my data for training?
Not necessarily. Memory controls and data-training controls are separate settings. Turning off memory affects personalization, not model improvement. Check the privacy and data controls section of your account settings if training use matters to you.
Why can’t I find the memory toggle in my settings?
The menu path has changed repeatedly and can vary by plan, app version, and region. Look for Personalization as well as Memory, since the label differs across versions. If neither appears on mobile, sign in on a browser, where the setting is usually available first, and it will sync to your account.
What is Temporary Chat and should I use it instead?
Temporary Chat creates a conversation that doesn’t appear in your history and isn’t used for memory. It’s ideal for one-off sensitive questions. The limitation is that it doesn’t sync across devices, so you can’t pick the conversation back up on another device later.
If I turn memory back on, does ChatGPT remember everything from before?
It can access memories that were saved and not deleted, plus past conversations if chat history reference is enabled. Anything you cleared is gone permanently. This is why clearing memories, rather than only toggling the switch, matters if you want a genuine reset.
