Switching from ChatGPT to Grok Bot: how to bring your context over with Import Memory (unofficial guide)

The short answer

Update Grok Bot to 0.75.2 or later, open Settings → General, and choose Start Import under Import. Copy the prompt Grok Bot gives you into a chat with ChatGPT, paste ChatGPT's answer into the second box, and choose Add to Memory.

Grok Bot sends the import to your primary Bot and opens that chat, so you can read what it picked up and correct anything that's off right away.

Memory covers preferences and context. Your old chat history, files, saved logins, and recurring jobs don't come over on their own; the sections below show where each one goes.

This guide is independent and unofficial. It isn't affiliated with or endorsed by xAI, Grok, Cursor, or OpenAI. Features change often, so check the linked docs and changelog.

Run Import Memory, step by step

Open Settings from the account menu, or press Cmd/Ctrl+, on desktop. Import sits in the General pane. If you don't see it, check that you're on Grok Bot 0.75.2 or later; the docs note that some settings depend on your account and rollout.

Choose Start Import. Grok Bot shows a prompt to copy. Paste it into a new chat with ChatGPT, or with whichever assistant you're leaving, such as Claude or Gemini.

Read the answer before you copy it back. It's a summary of what that assistant knows about you, so delete anything you don't want your Bot to keep, such as an old job, a client's name, or anything sensitive.

Paste the answer into the second box and choose Add to Memory. The import goes to your primary Bot, the main Bot that coordinates your other Bots. If you don't have one yet, create one first; a shared Bot can't be your primary Bot.

Check what your Bot learned

Personal Bots have no memory screen. In the chat that opens after the import, ask: "What do you remember about me?"

Fix mistakes in plain words: "That's out of date. I sold that business in 2025." Remove things the same way: "Forget that I'm on the launch team."

Each Bot has its own memory, but a few facts about you, such as your name and time zone, are shared by all your Bots. If a shared fact is wrong, tell any of your Bots the right version.

Resetting a conversation keeps memory. Deleting a Bot deletes that Bot's memory, so don't delete your primary Bot to "start fresh" unless you mean it.

What to bring over, and what to leave behind

Bring over how you like to work (tone, format, length), the rules you always repeat ("always ask before you send an email"), and stable facts: your role, your business, your time zone, and the tools you use.

Leave behind anything that changes often, like prices, pipelines, or this week's priorities. The docs say memory isn't a source of truth: keep changing facts in the source system, such as a spreadsheet or your CRM, and have the Bot reopen it when it matters.

Never put passwords, card numbers, or one-time codes in memory. Logins, cards, and contact info belong in Settings → Vault, which fills them in without showing them to the Bot.

Back important boundaries with approval rules, not memory alone. A rule in Settings → Permissions (Always Ask) gets an approval card every time; a remembered preference is only an instruction.

Move the rest of your ChatGPT setup

Saved prompts and custom instructions you reuse for one job: turn them into a skill. Paste the prompt and say "Save this as a skill called Weekly newsletter draft." See our guide to teaching a Bot a task with skills.

Tasks you ask for every week, like a Monday summary: turn them into a routine on a schedule instead of asking each time.

Files and reference docs: send them to the Bot in chat, or connect the app they live in (Google Drive, Notion, Gmail and others) from Connect apps in the sidebar, so the Bot reads the current version instead of a stale copy.

Different jobs: if you kept separate ChatGPT projects for separate areas, consider one focused Bot per long-lived job. The docs suggest starting small and adding a Bot only when the work has a stable specialist role.

Model choice: Grok Bot picks the model for you, and there's no model picker in Settings.