SecondBrain · Blog · ChatGPT as a second brain

How to use ChatGPT as a second brain, and what it still forgets

By Federico Cesarini, maker of SecondBrain · Updated October 29, 2026 · 8 min read

What ChatGPT remembers today

ChatGPT's memory is not one thing. As of October 2026 it is a handful of features that work together and behave differently. The main memory controls live under Settings, Personalization, Memory, and OpenAI notes that the exact options vary by plan, region and platform.

Two details matter for anyone hoping to use this as a second brain. First, OpenAI is explicit that memory does not keep every detail: it is the assistant's own working notes about you, not a record of what you told it. Second, there is no published size: the pricing page says memory is limited on Free and expanded on Plus and Pro, and that memories take up part of the model's context window.

Set up ChatGPT as a light second brain in 10 minutes

Within those limits, a short setup makes it much more useful.

1. Turn memory on. In Settings, Personalization, Memory, switch on saved memories and chat history if your plan offers them.
2. Write your custom instructions. Who you are, what you are working on this quarter, the people who matter in your work, and how you like answers. ChatGPT applies them across your chats, and a project's own instructions take priority inside that project, so keep both current.
3. Make one project per area of your life. Home, work, health, a side project. Put the relevant files in each. For a client or a topic you want kept apart, choose project-only memory in the project's settings, so other chats do not draw on it; OpenAI says the change can take a few hours.
4. Say what to keep, in plain words. "Remember that my daughter's swimming lessons moved to Tuesdays." Explicit requests become saved memories you can see and delete.
5. Talk instead of typing. Voice mode works on the web, iPhone and Android, which makes it far easier to offload a thought while walking.
6. Review once a month. Open the memory summary and saved memories, delete what is wrong or stale, and correct what drifted.

That gets you an assistant that knows your context. It does not yet get you a second brain.

Where ChatGPT falls short as a second brain

A second brain has three jobs: capture, organize and return. ChatGPT is good at the third and weak at the first two.

• It decides what to keep. You can ask it to remember something, but you cannot count on it to have kept a detail you mentioned in passing three weeks ago. A second brain you cannot trust gets abandoned.
• It is hard to browse. You can see and delete saved memories, but what it draws from chat history has no list, and OpenAI says project memory does not show one either. There are no folders of notes, so when you are not sure what to ask, there is little to look through.
• Capture lives inside chats. Every thought has to start a conversation, and quick notes end up scattered across dozens of threads.
• Dates do not file themselves. ChatGPT has scheduled tasks, capped at five active ones on Plus, and calendar plugins that can read and update your calendar when you ask. But a date you mention in passing stays in the chat unless you ask for an event.
• It belongs to ChatGPT. Claude and Gemini cannot read it. Claude can import a one-time snapshot, and Gemini offers an import too, though not in the European Economic Area, Switzerland or the UK, but nothing stays in sync. If you use more than one assistant, you are maintaining several memories by hand.

None of this is a flaw in ChatGPT. Its memory was built to make conversations feel continuous, not to be your archive, and expecting it to be both is where people get frustrated.

Give ChatGPT a second brain it can read

The setup that works is to keep your memory outside the chat, in a place you own and can browse, and let ChatGPT reach it. The bridge is the Model Context Protocol (MCP), the open standard that lets assistants call outside tools. ChatGPT supports MCP plugins on the web, but OpenAI's own pages currently disagree on which plans can add a custom one: the pricing page lists developer mode for Plus and Pro, while the help center describes full access on Business, Enterprise and Edu and read-only access on Pro. Check both for your plan before you set it up.

This is the job SecondBrain was built for:

• Capture anywhere. On your phone you tap and talk, for up to two minutes, and it writes the note, files it in the right folder and turns the events and reminders you mention into real ones. No chat to open.
• Browse and fix. Your notes sit in folders you can open, edit and export as Markdown (Pro, on iPhone).
• Ask from any assistant. Connected over MCP, ChatGPT can look things up in your SecondBrain when you mention it in a chat and, where your plan allows writing, save new notes after you confirm. Claude, Gemini, Cursor and Claude Code read and add to the same memory, so switching assistants no longer means starting over.

The shared memory is part of SecondBrain Pro. The full setup for each assistant, including Codex and ChatGPT Work, which install it from the same plugin marketplace as Claude, is in how to share one memory between ChatGPT and Claude.

Prompts that help it recall

Once ChatGPT can read a real memory, the way you ask changes. These work best with a memory connected; the save prompt also needs a plan that lets ChatGPT write to it:

• To decide: "What have I already decided about the new apartment, and what is still open?"
• To prepare: "I'm meeting Sara from the design agency tomorrow. What do I know about her and our last conversation?"
• To save: "Save this in my SecondBrain: the boiler service is booked for November 14 at 9, and the technician calls half an hour before."
• To clean up: "List everything you remember about my work projects, grouped by project, and flag anything that looks out of date."
• To write: "Draft the update email for the team using my notes on the Q4 launch."

Asking ChatGPT to list what it remembers is also the fastest way to audit its memory, and it is the same idea behind the export prompt that Claude's import asks you to run.

What to keep out of it

Whatever you use, a second brain will end up holding your life, so decide early what goes in. Use temporary chats for anything you do not want remembered; OpenAI says they create no memories while they stay temporary, though it may keep a copy for up to 30 days. If you do not want your chats used to train models, check the Improve the model for everyone setting in ChatGPT's data controls. Keep passwords, ID numbers and anything you would not email to yourself out of every AI tool. And know where your notes live: SecondBrain, for instance, stores them in the EU, processes voice and text with Google's paid Gemini API, and never uses your notes to train models.

Questions people ask