Make ChatGPT, Claude and Grok all know you: move your AI memory and keep a copy you own

1. Workflow

1. Ask the chatbot you use most to export everything it remembers about you in one chat, leaving out topics you don't want to move. Save the result in a text file.

   Tool

   Your current chatbot (ChatGPT, Grok or Claude)

   Why

   Leaving Claude? Start by asking Claude: "Write out your memories of me verbatim, exactly as they appear in your memory," and save the reply (Claude Help Center, checked Oct 20, 2026). Leaving ChatGPT? OpenAI's help center says you can ask ChatGPT what it remembers about you, and that the memory summary under Settings > Personalization > Memory doesn't necessarily include every detail, so ask in chat rather than copying the summary (checked Oct 20, 2026). Grok has no documented memory export. For any of the three, our prompt below follows Anthropic's recommended export prompt (every memory in one list, dates where available, your own words, and a check on whether the list is complete) and adds a short interview so you can leave out sensitive topics first.

   Cost

   $0 on free plans.

   Copy prompt Paste into the chatbot you're moving from (ChatGPT, Grok or Claude).

   Preview prompt: Export my memory (with an opt-out interview)

   ROLE You are a careful data-export assistant. You help me take everything you remember about me out of this chatbot so I can keep my own copy and give it to another assistant. GOAL Produce one complete, honest export of the memories and context you hold about me, minus anything I choose to leave out. WHAT THE NEXT STEP NEEDS A raw export for Step 2: every stored memory or learned piece of context, one entry per line, with the date saved if you have it, in my own words where possible. HOW TO WORK WITH ME Do not write the export yet. Interview me first. - Ask your questions ONE AT A TIME and wait for my answer before asking the next. (If I ask for speed, you may ask a short numbered batch of at most 4 questions instead.) - With every question, offer 2-3 sensible example answers or a default, so I can reply "default" or pick one. - If an answer is vague, ask one follow-up. If I don't know, choose a sensible default and tell me what you chose. - Do not generate the export until I have answered every question. When you have enough, give me a 3-5 line summary of what you understood and ask me to confirm. - Do not invent memories. Only list what you actually have stored or have learned from our past chats. If you are unsure whether something is stored, say so next to that entry. QUESTIONS TO COVER (in a sensible order, skip any I already answered above) 1. Where is this export going? Examples: Claude's memory import, ChatGPT custom instructions, Grok custom instructions, just a backup file. (Default: a backup file I keep, then Claude's import.) 2. Which topics should be left out entirely? Examples: health, money, family members' details, relationships, anything about clients. (Default: leave out health, finances and other people's personal details.) 3. Should work and personal context both be included, or work only? (Default: both, clearly labelled.) 4. Do you want dates on each entry where you have them? (Default: yes.) THE DELIVERABLE (only after I confirm) One code block with every remaining entry, formatted as: [date saved, if available] - memory content. Cover all of these, keeping my words verbatim where possible: 1. Instructions I've given you about how to respond (tone, format, style, "always do X", "never do Y"). 2. Personal details I chose to keep: name, location, job, interests. 3. Projects, goals and recurring topics. 4. Tools, languages and apps I use. 5. Preferences and corrections I've made to your behaviour. 6. Any other stored context not covered above. Do not summarize, group or drop entries, except the topics I asked you to leave out. After the code block, tell me plainly whether this is the complete set or whether some memories remain that you could not list. Put the export inside one clearly delimited block that starts with a line "=== COPY START ===" and ends with a line "=== COPY END ===", with no commentary inside the block. Put any notes outside it. AFTER THE DELIVERABLE Ask me: "Anything to tweak?" If I ask for changes, revise and give me the full updated block again between the same markers. Repeat until I say it is good. FINISH When I'm happy, end with one line telling me what to do next: save this export in a text file, then go to Step 2 to turn it into your own portable memory file.

2. Paste the export into any chatbot and turn it into your own about-me.md file: a full memory file plus a short version under 1,000 characters. Save it on your computer.

   Tool

   Any chat (Claude, ChatGPT or Grok)

   Why

   Memory inside one app is hard to move. A plain file you own works with every assistant, and OpenAI's help center says ChatGPT's memory doesn't keep every detail from every conversation, and that custom instructions are separate, direct guidance you write yourself (checked Oct 20, 2026). Cleaning it first also removes outdated or sensitive entries before they reach another app.

   Cost

   $0 on free plans.

   Copy prompt Paste into ChatGPT, Claude or Grok.

   Preview prompt: Build my portable about-me file

   ROLE You are an editor who turns a messy chatbot memory export into a clean, portable "About me" file that any AI assistant can use. GOAL Help me build one memory file I own, plus a short version that fits in a custom-instructions box. WHAT THE NEXT STEP NEEDS Two texts for Steps 3 to 5: a FULL MEMORY FILE to import into Claude and keep as my backup, and a SHORT VERSION to paste into ChatGPT's and Grok's custom instructions. HOW TO WORK WITH ME Do not write the deliverable yet. First ask me to paste my export(s) from Step 1. Then interview me. - Ask your questions ONE AT A TIME and wait for my answer before asking the next. (If I ask for speed, you may ask a short numbered batch of at most 5 questions instead.) - With every question, offer 2-3 sensible example answers or a default, so I can reply "default" or pick one. - If an answer is vague, ask one follow-up. If I don't know, choose a sensible default and tell me what you chose. - Do not generate the deliverable until I have answered every question. When you have enough, give me a 3-5 line summary of what you understood and ask me to confirm. - Do not invent facts about me, my work or my preferences. Only use what is in my export or what I tell you. If two entries conflict, ask me which is right. QUESTIONS TO COVER (in a sensible order, skip any I already answered above) 1. Which entries are out of date or wrong? (Default: you list anything that looks old or conflicting and I confirm.) 2. Is there anything sensitive left that should come out before this goes into another app? Examples: health, money, other people's names, client details, passwords. (Default: remove all of those.) 3. What should every assistant know first about your work? Examples: your role, your main project this month, the audience you write for. (Default: role plus your top 2 current projects.) 4. How do you like answers? Examples: short with bullet points, step by step, plain English with no jargon. (Default: short, direct, bullet points when it helps.) 5. What should an assistant never do? Examples: invent facts, use emojis, write long intros, change my code style. (Default: never invent facts; ask when unsure.) THE DELIVERABLE (only after I confirm) 1. FULL MEMORY FILE with these headings: About me; How to respond; Current projects and goals; Tools and apps I use; Always / Never; Last updated (today's date). One fact per line, no duplicates. 2. SHORT VERSION: the most important facts and response preferences only, under 1,000 characters, written as instructions an assistant can follow (for example "I'm a ... Keep answers ... Never ..."). Write in plain English. Keep my own wording where it was clear. Put everything I should copy inside one clearly delimited block that starts with a line "=== COPY START ===" and ends with a line "=== COPY END ===", with no commentary inside the block. Label the two parts inside the block. Put any notes outside it. AFTER THE DELIVERABLE Ask me: "Anything to tweak?" If I ask for changes, revise and give me the full updated block again between the same markers. Repeat until I say it is good. FINISH When I'm happy, end with one line telling me what to do next: save both parts as a file called about-me.md, then go to Step 3 to import the full file into Claude.

3. Import the full file into Claude: Settings > Memory > Start import, paste the FULL MEMORY FILE, then click Add to memory. Use See what Claude learned about you to check the result. If you only see Memory under Settings > Capabilities, you're on the legacy experience and the screens differ.

   Tool

   Claude (web or Claude Desktop)

   Why

   Anthropic says Claude extracts key information from the pasted text and stores it as individual memory entries you can review in the memory panel. Imports are experimental, so the next step checks what actually stuck.

   Cost

   $0: memory import is listed for Free, Pro, Max, Team and Enterprise plans.

4. In a new Claude chat, compare what Claude kept with your file, approve fixes, and note anything Claude won't store so it can go into your short version instead.

   Tool

   Claude

   Why

   Anthropic says Claude's memory focuses on work-related topics, so personal details unrelated to work may not be kept, and that you can review and change what Claude remembers, including by asking Claude in chat.

   Cost

   $0 on Claude Free.

   Copy prompt Paste into a new Claude chat.

   Preview prompt: Check and fix what Claude remembered

   ROLE You are Claude, and you are checking how well my memory import worked against the file I keep myself. GOAL Find what you kept, what you missed and what you got wrong from my imported memory, and fix it with my approval. WHAT THE NEXT STEP NEEDS A clean, confirmed memory in Claude that matches my about-me file, and a short list of anything Claude would not keep, so I can put it in Step 5's short version instead. HOW TO WORK WITH ME Do not change anything yet. First ask me to paste my FULL MEMORY FILE from Step 2. Then interview me. - Ask your questions ONE AT A TIME and wait for my answer before asking the next. (If I ask for speed, you may ask a short numbered batch of at most 4 questions instead.) - With every question, offer 2-3 sensible example answers or a default, so I can reply "default" or pick one. - If an answer is vague, ask one follow-up. If I don't know, choose a sensible default and tell me what you chose. - Do not change or add memories until I have answered every question. When you have enough, give me a 3-5 line summary of what you plan to change and ask me to confirm. - Do not invent memories or claim you stored something you did not. If you are unsure whether something is in your memory, say so. QUESTIONS TO COVER (in a sensible order, skip any I already answered above) 1. Compare my file with what you now remember about me. Show me three short lists: kept, missing, and different. Then ask: which missing items matter? (Default: everything under "How to respond" and "Current projects and goals".) 2. Which of the "different" items is right, my file or your memory? (Default: my file is right.) 3. Are there personal, non-work details you won't keep that I should move to the short version instead? (Default: yes, list them for me.) 4. Is there anything you remember that should be removed? (Default: remove anything older than my file's "Last updated" date that conflicts with it.) THE DELIVERABLE (only after I confirm) 1. Make the agreed memory changes, then tell me plainly which ones you could and could not apply. 2. CHANGE LOG: added, corrected and removed items, one per line. 3. NOT KEPT: items Claude didn't store, so I can add them to my short version. Put the change log and the not-kept list inside one clearly delimited block that starts with a line "=== COPY START ===" and ends with a line "=== COPY END ===", with no commentary inside the block. Put any notes outside it. AFTER THE DELIVERABLE Ask me: "Anything to tweak?" If I ask for changes, revise and give me the full updated block again between the same markers. Repeat until I say it is good. FINISH When I'm happy, end with one line telling me what to do next: add the NOT KEPT items to your short version if they matter, then go to Step 5 to give ChatGPT and Grok the short version.

5. Give ChatGPT and Grok the short version. In ChatGPT on web or desktop, open Settings > Personalization, make sure Enable customization is on, and paste it into the Custom Instructions field (on iOS and Android the menu is called Customize ChatGPT). In Grok, paste it into Grok's custom instructions or personalization settings; xAI doesn't publish the menu path, so the wording in your app may differ. Then run the test prompt in each to check it was understood.

   Tool

   ChatGPT and Grok

   Why

   OpenAI's help center says custom instructions apply to all your chats and hold up to 1,500 characters on Free and Go, or 5,000 on paid plans, so the short version fits (checked Oct 20, 2026). xAI's Grok page lists custom instructions and memory across chats as Grok features but gives no menu path. Neither company documents a memory import like Claude's on its official pages, so custom instructions are the reliable route.

   Cost

   $0 on ChatGPT Free and Grok Free.

   Copy prompt Paste into ChatGPT, then into Grok, after saving your custom instructions.

   Preview prompt: Test and tighten my custom instructions

   ROLE You are the assistant I just gave my custom instructions to. You are helping me test that you understood them and tighten the wording. GOAL Check that my short "About me" version works in this assistant, and improve it if anything is missing or misread. WHAT THE NEXT STEP NEEDS A tested SHORT VERSION I can keep in my about-me.md file for Step 6 and paste into any other assistant. HOW TO WORK WITH ME Do not rewrite anything yet. First tell me, in 5 lines or fewer, what you know about me and how I want answers, based only on my custom instructions and any memory you have. Then interview me. - Ask your questions ONE AT A TIME and wait for my answer before asking the next. (If I ask for speed, you may ask a short numbered batch of at most 4 questions instead.) - With every question, offer 2-3 sensible example answers or a default, so I can reply "default" or pick one. - If an answer is vague, ask one follow-up. If I don't know, choose a sensible default and tell me what you chose. - Do not generate the deliverable until I have answered every question. When you have enough, give me a 3-5 line summary of what you understood and ask me to confirm. - Do not invent facts about me. If something is not in my instructions or your memory, say it isn't there. QUESTIONS TO COVER (in a sensible order, skip any I already answered above) 1. Is anything in your 5-line summary wrong or missing? (Default: you point out anything my instructions leave unclear.) 2. Which recurring task should we test with? Examples: draft a reply to a client email, plan my week, explain a news story for my audience. (Default: a short email draft.) 3. After you do that test task, did it match my style? What should change? (Default: you suggest one change.) 4. Should anything be cut to keep the instructions short? (Default: cut anything you didn't need for the test.) THE DELIVERABLE (only after I confirm) 1. The test task output (short). 2. An improved SHORT VERSION of my custom instructions, under 1,000 characters, in plain English. 3. One line on what changed and why. Put the improved short version inside one clearly delimited block that starts with a line "=== COPY START ===" and ends with a line "=== COPY END ===", with no commentary inside the block. Put any notes outside it. AFTER THE DELIVERABLE Ask me: "Anything to tweak?" If I ask for changes, revise and give me the full updated block again between the same markers. Repeat until I say it is good. FINISH When I'm happy, end with one line telling me what to do next: paste the improved version back into this assistant's custom instructions and into your about-me.md file, then go to Step 6 to keep it current.

6. Keep about-me.md as the master copy. Once a month, re-run Step 1 in each assistant, merge anything new into the file, and update the custom instructions. For chats you don't want remembered in Claude, use an incognito chat.

   Tool

   Your about-me.md file

   Why

   Each assistant keeps learning separately, so the copies drift apart. Anthropic says incognito chats are available on all Claude plans and that you can turn off Search and reference chats in Settings > Memory.

   Cost

   $0.

2. Recommended tool chain

1. Your current chatbot (ChatGPT, Grok or Claude)
2. Any chat (Claude, ChatGPT or Grok)
3. Claude (web or Claude Desktop)
4. Claude
5. ChatGPT and Grok
6. Your about-me.md file

3. Total cost

Unpriced items are left out of the total (marked “cost not sourced”), so the real total is higher. All totals are our arithmetic from sourced unit prices.

4. Current promos

• Claude Pro annual billing: $17/month vs $20 monthly

  Anthropic's pricing page lists Pro at $17/month with an annual subscription ($200 billed up front) versus $20/month billed monthly. This is a standing discount, not a limited-time sale.

  Applies to: Claude Pro on the best route ($17/month billed annually vs $20 monthly)

  Standing, no end date. Only worth it if you'll keep Pro for a year.

  Source: Anthropic: Claude pricing, checked Oct 20, 2026

• ChatGPT Plus: 4 months free for US college students (ends Oct 31, 2026)

  OpenAI's student page offers eligible US college students four free months of ChatGPT Plus, including access to ChatGPT Work (an $80 value), verified through SheerID. Claim by October 31.

  Applies to: ChatGPT in Step 5 (not needed for this playbook; Free works)

  Renews at $20/month after the free period unless you cancel. Not available for App Store or Google Play billing.

  Source: OpenAI: ChatGPT for students, checked Oct 20, 2026

5. Learn it

Last verified Oct 20, 2026. Items older than 30 days are marked.

6. Caveats and assumptions

• Claude says memory imports are experimental and may not always incorporate everything, and that its memory focuses on work-related topics, so personal details may be dropped. Step 4 checks this.
• An export is only what the chatbot says it remembers. It may miss things or phrase them differently, which is why the export prompt asks it to say whether the list is complete.
• We found no memory import for ChatGPT or Grok on their official pages (checked Oct 20, 2026). ChatGPT steps follow OpenAI's help center, checked Oct 20, 2026: its data export (Settings > Data controls > Export data) can take up to 7 days, and uploading exported conversations to another ChatGPT account doesn't transfer custom instructions, memories or other settings. xAI's Grok page lists memory across chats and custom instructions but gives no menu path, so we don't name Grok menu labels; look in Grok's settings, as the wording may differ in your app.
• Claude's help page names two different places to view your memory (Settings > Capabilities in one section, Settings > Memory > Topics in another; checked Oct 20, 2026), so for getting memory out of Claude we lead with asking Claude to write out its memories verbatim, which doesn't depend on either menu.
• Take out health, money, passwords and other people's personal details before pasting your export into any app. Anything you import becomes part of that company's stored data under its own policies.
• If your Claude settings show Memory under Settings > Capabilities, you're on the legacy memory experience and some screens differ.
• Plan features and prices are from each company's pages on the verified date and can change; costs are our arithmetic, one month unless stated.

Last verified Oct 20, 2026.