folk
skills
characters
faq
earn
log inget the app
get the app
← blogguides

How to Move Your ChatGPT Memory to a Texting AI Assistant

Switching from ChatGPT or Claude doesn't mean starting over. Here's how to move your saved memory into a texting AI assistant in one paste.

folk teamAugust 14, 2026
How to Move Your ChatGPT Memory to a Texting AI Assistant

on this page

Why does switching AI assistants feel like starting over?How do you get your saved memories out of ChatGPT?How do you get your saved memories out of Claude?What's the fastest way to copy what an AI remembers about you?How do you move that memory into a texting AI assistant?How folk imports your ChatGPT or Claude memoryThe bottom line

You've spent months telling ChatGPT or Claude who you are. Your job, your partner's allergies, the trip you're saving for, how you like your emails written. Then you find an AI assistant that actually lives in your texts, one that can remind you about things and message you first instead of waiting quietly in a browser tab, and the whole idea stalls on one question: do you have to teach it everything again?

You don't. What an assistant remembers about you isn't welded into it. It's just text, written in plain language, and text can move.

This guide covers how to get your saved memory out of ChatGPT or Claude, what actually survives the move, and how folk handles the import on its end.

Quick answer: Ask ChatGPT or Claude to write out everything it has saved about you in one message, then paste that reply into folk's memory import page (or upload your ChatGPT data export directly). folk reads it, folds duplicate facts together, and adds what survives to its own long-term memory, so a brand-new folk account can already know your basics instead of starting from zero.

Why does switching AI assistants feel like starting over?

Because memory is not portable by default. Every assistant stores what it learns about you inside its own account and its own systems, and none of them hand that off to a competitor automatically. Nobody designed it to be hard to leave, it's just that nobody built the door either, until recently.

The upside: since memory is just text the assistant can read back to you, you don't need an official export feature to move it. You need the assistant to say it out loud, once, in a format you can copy.

How do you get your saved memories out of ChatGPT?

Two ways, depending on how much you want:

  1. The official data export. Open Settings, then Data Controls, then Export, and confirm. OpenAI emails you a download link within a few days, and the OpenAI Help Center notes the link expires 24 hours after it arrives, so grab it promptly. This is the fullest record, since it includes your saved memories and any custom instructions you've set, in a structured file rather than a summary ChatGPT writes on the spot.
  2. Just ask it. Open a chat and ask ChatGPT to list everything it remembers about you. It reads back its own saved memory in a message you can copy straight away, no waiting on an email.

The export is more complete. Asking directly is faster if you just want the basics moved today.

How do you get your saved memories out of Claude?

Claude doesn't offer a separate memory export file. What you get instead is a Settings page where you can view and edit individual memory entries, and the same trick that works for ChatGPT: ask Claude, in a normal message, to write out everything it has saved about you.

This isn't a workaround anyone had to invent. Anthropic ships nearly this exact copy-paste flow in the other direction, a prompt that pulls a user's memory out of ChatGPT and into Claude, which MacRumors reported when Anthropic opened memory up to free Claude users. If one of the model makers builds a paste-based import as the official way to switch assistants, it's a reasonable way to do it in any direction.

What's the fastest way to copy what an AI remembers about you?

Ask for it in a structure you can actually use, not just a paragraph. Something like:

Write out everything you know about me from the saved memories and conversations you can actually access. Include both things I've told you and things you've picked up from our conversations. Keep my own words wherever possible, especially for instructions and preferences; quote them exactly when available.

Use these six category headings, in this order:
1. Instructions: Rules I've asked you to follow about tone, format, style, always/never behaviors, and corrections I've made to your answers. Attribute each rule to me and make its target clear, for example: I asked you to "keep answers short". Do not turn a correction aimed at you into a fact about me.
2. Identity: Name, age, location, education, family, relationships, languages, personal interests, and relevant personal circumstances, including dietary needs, routines, schedules, and constraints.
3. Career: Current and past roles, companies, career history, and general skill areas.
4. Projects: Projects I've meaningfully built or committed to. Start each entry's content with the project name. Include what it does, its current status, key decisions, constraints, and next steps where known. Prefer one entry per project, but use additional entries when needed to preserve all details.
5. Goals: Things I'm working toward or have said I want, including their current status, milestones, deadlines, and blockers where known.
6. Preferences: Broad opinions, tastes, habits, and working-style preferences. Include anything else worth remembering in the most relevant category rather than leaving it out.

Accuracy and completeness:
- Preserve all distinct details. Do not replace specifics with a broad summary, merge unrelated facts, or leave something out merely because it seems minor. Consolidate only exact repeats without losing detail.
- Clearly distinguish things I told you from things you worked out. End every inferred entry with (inferred); include the basis or uncertainty when relevant. Do not invent facts or fill gaps.
- Preserve changes over time and mark outdated or superseded information. If facts conflict and you cannot resolve them, say so rather than choosing silently.
- Only use memories and conversation history available to you. Do not imply you searched chats you cannot access.

Format:
- Put the export in a single code block, with no introduction.
- The first line inside the code block must be "From: [the name of the assistant writing this]", replacing the bracketed placeholder with your name.
- Use a Markdown heading for each category, such as "## Instructions". Keep all six headings; leave a category empty if nothing is known.
- Within each category, put one entry per line, oldest first where dates are known. Use: [YYYY-MM-DD] - Entry content here.
- If the date is unknown, omit it and use: - Entry content here. Never invent a date or use today's date for an undated memory. If only a rough date is known, mention it in the entry instead.
- Keep each entry under 1,000 characters so it imports without being cut off. Split longer entries into additional self-contained lines, repeating the project name where applicable.

After the code block, state whether this is everything available to you or whether there is more. Mention any inaccessible history or other limits on completeness. If the response limit prevents a complete export, stop at a complete entry, say what remains, and ask me to continue; do not claim the export is complete.

That last line matters. Assistants sometimes hold back context they consider not worth surfacing, so asking whether anything was left out at least tells you the copy is incomplete rather than letting you assume it's everything.

How do you move that memory into a texting AI assistant?

Once you have the reply copied, paste the whole thing into the new assistant's import flow if it has one. folk's import memory page is built for exactly this: paste the reply, or upload your raw ChatGPT export if you did the fuller version, and it parses either one into individual memory entries you can review before anything saves.

You'll see the entries form as you paste, near-duplicate lines fold into one, and a short status line tells you how many are ready to save. Review it, trim anything that doesn't belong, and save. If you're not signed up yet, folk holds the parsed batch until you finish creating an account, then adds it to your memory automatically.

How folk imports your ChatGPT or Claude memory

folk is the personal AI assistant that lives in your texts (iMessage, Telegram, and WhatsApp), and importing memory from somewhere else is meant to be a five-minute step, not a project:

  • Two ways in. Paste the copied reply from ChatGPT or Claude directly, or upload your ChatGPT data export if you want the fuller version that includes custom instructions. See memory for how folk's memory works day to day.
  • Duplicates fold together automatically. If ChatGPT and Claude both remember you're vegetarian, that shows up once, not twice.
  • You review before it saves. Nothing is added to your memory until you confirm the batch, so you can cut anything you'd rather not bring over.
  • It becomes real memory, not a one-time note. Once saved, the imported facts sit alongside everything folk learns from talking to you going forward, in the same long-term memory it uses across iMessage and Telegram (WhatsApp is rolling out by region).
  • Encrypted at rest, never used to train models. Imported memory is encrypted with AES-256-GCM the moment it reaches the server, and folk does not train on your data.

Because it's the same memory no matter which app you text, you only have to do the import once.

The bottom line

The context you've built up in ChatGPT or Claude isn't stuck there. It's a conversation you can ask for and a paste away from wherever you want it next. If you're ready to try a texting AI assistant, bring what you already have with you at folk's memory import page, then get started with folk from a place that already knows the basics instead of a blank one.

If dates and people are what you most want it to hold onto, see how to remember birthdays and important dates. For the bigger case on why a texting assistant is worth switching to in the first place, read the best AI assistant that lives in your texts. And if you're still deciding between Claude and ChatGPT and want to know how their memory features actually differ before you export anything, see does Claude have memory like ChatGPT?

meet folk

The personal AI that lives in your texts - iMessage, Telegram, and WhatsApp. Free to start.

start free

related guides

  • guidesAI Characters You Can Text on iMessage (Gojo, Levi, Goggins)AI characters you can text on iMessage: how character texting works, why it beats an app, which characters people pick, and how to make your own.
  • guidesBest Morning Brief Apps and Flash Briefing Alternatives (2026)Alexa Flash Briefing, Alexa+, Gemini Daily Brief, and newsletter digests compared, plus a texting assistant that sends one in your normal chat.
  • guidesHow to Stop a Group Chat From Re-Asking Scheduling QuestionsHow to keep a group chat from re-asking the same scheduling question for a recurring plan, using shared memory instead of a new poll each time.
folkfolk status

made with by Nozomio Labs© 2026

folk status

company

  • Nozomio Labs
  • careers
  • privacy
  • terms
  • subprocessors
  • status

explore

  • characters
  • skills
  • faq
  • use cases
  • blog
  • compare

where

  • iMessage
  • Telegram
  • WhatsApp

connect

  • email
  • TikTok
  • Instagram
  • X / Twitter

folk