AI conversations

How to move a conversation from ChatGPT to Claude or Gemini

Updated October 14, 2026 · 3 min read

There are two reasons to move a conversation from ChatGPT to Claude, or in any other direction. Either you are switching assistants for good and want your history to follow, or you have one conversation that stalled and want a second opinion without retyping the context. The official import tools address the first case. For the second, a Markdown transcript is quicker.

What the official import tools do

Both Anthropic and Google shipped import features in 2026. They are built for people changing assistants, and they work at account level.

Claude: Import Memory. PCWorld reported on March 3, 2026 that Claude can bring over what ChatGPT, Gemini or Copilot know about you. The tool produces a prompt, you run it in your old assistant, and you give the result to Claude. According to the same article, Claude takes about 24 hours to absorb the memories, and you can review and edit what was saved.

Gemini: import memory or chat history. PCWorld reported on March 27, 2026 that Gemini offers two options under Settings: pasting a memory summary from your old assistant, or uploading the ZIP export of your ChatGPT or Claude history so you can continue old conversations in Gemini.

Both are worth using when you migrate. Neither fits the everyday case:

  • They act on everything at once. You cannot pick one conversation.
  • A memory import is a summary of you, not the content of a discussion.
  • A full-history import starts with an account export, which arrives by email.
  • Availability depends on your plan and region, so check your own settings.

Moving one conversation: the Markdown transcript

A conversation is text. Give the new assistant that text, clearly labeled, and it has everything the old one had.

  1. Open the conversation in Chrome and press Alt+M to open Markforge. It has a dedicated parser for ChatGPT, Claude, Gemini and Grok.
  2. Read the preview. Each turn sits under a User or Assistant heading.
  3. Trim. The content field is editable, so delete the dead ends and keep the exchange that matters.
  4. Click Copy.
  5. Paste into a new chat in the other assistant, with a line of instruction above it.

An instruction that works:

Below is a conversation I had with another assistant. Read it, then
continue from the last message. Point out anything in the earlier
answers you disagree with.

The last sentence earns its place. A second model reading the first model's answers is a cheap way to get them reviewed.

If you would rather attach a file than paste, click Download .md and upload the file where the assistant accepts text attachments.

Will it fit?

Long conversations can outgrow the context you want to spend on them. The Markforge popup shows a token count under the content, next to characters and words.

Two things to know about that number. It is computed with a GPT tokenizer, so it is exact for OpenAI models and approximate for Claude or Gemini. And it updates as you edit, so you can watch it drop while you trim.

One toggle helps on link-heavy conversations. Strip links removes URLs and keeps the link text. On pages full of links the saving is real: we measured between 21% and 39% fewer tokens on four link-heavy pages, and about 2% on pages with few links.

What does not travel

A transcript is the visible conversation, and only that.

  • Memory and custom instructions. The old assistant's saved knowledge of you is not in the chat. Use the import tools above for that.
  • Files and images. Uploads, generated images and artifacts stay behind. Attach again what matters.
  • Reasoning. Claude's extended thinking sections are removed from the export. The final answers are kept.
  • Tables. In version 1.0.0, Markforge flattens tables to lines of text. If a table carries the discussion, copy that reply with the assistant's own copy button and paste it in place.
  • Tool results. Web searches or code the assistant ran appear only as far as their output was shown in the conversation.

Per-assistant details

Each export has its own quirks, covered in a dedicated guide:

Frequently asked questions

Can I import a single ChatGPT conversation into Claude?

Not with the import tools, which work on your memory or your whole history. The practical way is to export the conversation as Markdown, paste it into a new Claude chat, and ask Claude to continue from the last message.

Will the new assistant understand who said what?

Yes, if the transcript labels the turns. A Markforge export puts each turn under a User or Assistant heading, which models read without trouble. Add one line of instruction saying the text is a previous conversation.

How do I know if the conversation is too long to paste?

Check the token count in the Markforge popup before copying. It uses a GPT tokenizer, so it is exact for OpenAI models and an estimate for others. If it is large, trim the transcript in the editable content field.

Does the other assistant get my uploaded files and images?

No. A transcript carries text. Files you uploaded, images the assistant generated and artifacts are not part of it, so attach again whatever the next assistant needs.

Related guides

How to copy a web page as Markdown for ChatGPT or ClaudePasting a page into a model works better as Markdown than as HTML or raw text. Three ways to get it, and which one works on pages behind a login.How to export a ChatGPT conversation to MarkdownChatGPT can copy one reply as Markdown and export your whole account as JSON and HTML. Getting one conversation into one .md file takes a third route.How to export a Claude conversation to MarkdownClaude has no Export as Markdown button. The copy button, the official data export and a browser extension each solve a different part of the problem.How to export a Gemini chat to MarkdownGemini sends single replies to Docs or Gmail and your history to Takeout. Getting a whole conversation into one Markdown file takes one more step.