How to turn on memory in ChatGPT on Mac: steps, limits, and what to check

What you need before you turn memory on

Memory belongs to your ChatGPT account, so sign in first. The ChatGPT home page says you need to sign in to save chats, and memory needs a signed-in account too. If you use ChatGPT on the web and on your Mac, use the same account in both. Otherwise you will compare two different sets of settings and wonder why they do not match.

Your plan sets how much memory and context ChatGPT uses. The pricing page gives each plan a short label. Free gets limited memory. Go gets longer memory. Plus gets expanded memory and context. Pro gets maximum memory and context.

Guides disagree about which plans get memory. Some say only Plus, Team, and Enterprise users have it. One reviewer found it on every tier. The pricing page lists memory for Free, Go, Plus, and Pro, so use your own Settings screen as the final check. If the switch is missing, look at your plan and account first.

How memory works in the ChatGPT Mac app

Memory lets ChatGPT carry details from one chat into later chats. You do not have to repeat your job, your tools, or your writing style every time. That saves time on repeat questions. It also means ChatGPT can bring up an old detail at the wrong moment, so you need to read what it keeps.

An arXiv paper called The Algorithmic Self-Portrait looked at 2,050 memory entries from ChatGPT users. The authors report that 96% of the memories were created unilaterally by the conversational system. In plain words, the user did not ask for most of them. Keep that number in mind when you review what ChatGPT has saved.

Turn on memory step by step

The steps above list the clicks in order. This section explains what each click is for and what to look for on your screen. Keep the Settings window open while you follow along. Labels are the part most likely to differ between builds.

Open Settings with ⌘,

Settings opens with ⌘, from the app menu. You can also choose ChatGPT > Settings from the menu bar. Both routes open the same window. If ⌘, does nothing, click ChatGPT in the menu bar first so the app is active. Then press the shortcut again.

Find Personalization in the sidebar

Select Personalization in the Settings sidebar. On the web, OpenAI's Learn pages say ChatGPT memory is managed under Settings > Personalization. The guides we read describe the Mac app the same way. We could not open OpenAI's Mac help pages, so confirm the layout in your copy of the app.

Turn on the memory switch

Find the memory switch and turn it on. Third-party guides name two switches, Reference saved memories and Reference chat history. Saved memories are the notes ChatGPT keeps. Chat history lets it use earlier chats. If your build shows both, decide which one you want before you switch them on.

Confirm that memory works

Start a new chat. Ask a question that depends on something you told ChatGPT before, such as your job or a project name. Or ask what it knows about you. If the answer uses earlier details, memory is working. If it does not, check the switches again, then sign out and sign back in.

Check the default before you change anything

Sources disagree on whether memory starts on or off. One guide says it is off. A TechRadar reviewer also found it off. Accounts may differ, so look at the switch before you touch it. If it is already on, you may not need to do anything. If it is off, switch it on and run the test above.

Temporary Chat and memory

Temporary Chat is a separate option. Press ⌘⇧N to start one in the Mac app. OpenAI's commands reference lists it as New temporary chat, and the shortcut applies to ChatGPT, not Codex. Whether a temporary chat ignores memory is less clear. One OpenAI snippet suggests it does. Some guides say a temporary chat can read memories without creating new ones. We could not confirm either claim, so test it in your own account.

For the full steps, read how to use temporary chat in ChatGPT on Mac.

Mistakes people make with memory

• Turning on the Codex toggle and thinking memory is on. Those are two different settings.
• Trusting a guide's menu names without checking the window in front of you.
• Forgetting that saved entries apply across chats, so old details can appear in new ones.
• Skipping the test. A switch that looks right does not prove that memory works.
• Assuming every saved entry is accurate. The study found most entries came from the system, so check them.
• Treating memory as the only privacy control. Temporary Chat and chat history are separate choices.

These mistakes share one cause. People trust the first label they see. Open the switch, read the label, and test with a real question.

When memory seems to do nothing

Start with sign-in. Memory needs a signed-in account, so a logged-out window will not help. Then check your plan on the pricing page. Free gets limited memory, so less carries over from older chats. Last, check that you are not in a Temporary Chat, because that may not behave like a normal chat.

Some third-party blogs say memory fills up at a set limit. Those limits came from sources we could not check, so do not treat any number as official. OpenAI's pricing page only uses words like limited, longer, expanded, and maximum.

Turn memory off again

Use the same path. Open Settings from the menu bar, select Personalization, and switch memory off. The guides describe chat history as a separate switch, so check that one too if you want less saved. For those steps, read how to turn off chat history in ChatGPT on Mac.

Next steps after memory is on

Custom instructions also shape how ChatGPT answers you. Set them if you want a standing set of preferences. See how to set custom instructions in ChatGPT on Mac. If you share a Mac or a work account, read how to sign out of ChatGPT on Mac as well.

What the research behind the 96% figure does and does not show

The study used chat logs donated through European data access requests. That makes it a sample, not a survey of all ChatGPT users. Its user count also differs between the paper and a news report. Read the numbers as a rough picture, not a precise rate.