How do I read conversations.json from a Claude export?
It is one JSON array with a "chat_messages" list per conversation, and the readable text is inside each message's "content" blocks.
Updated November 15, 2026
What is in the export
On claude.ai, go to Settings → Privacy → Export data. The ZIP you get by email holds three files:
- conversations.json: every conversation, in one array.
- projects.json: your Projects, with their instructions and knowledge files.
- users.json: your account record.
How conversations.json is laid out
- Each item in the array is one conversation:
uuid,name(the title, sometimes empty),created_at,updated_atandchat_messages. - Each item in
chat_messagesis one message:senderis"human"(you) or"assistant"(Claude), pluscreated_at,text,content,attachmentsandfiles. contentis a list of blocks. A block with"type": "text"holds the words in itstextfield. Other block types arethinking,tool_useandtool_result.- An artifact is a
tool_useblock namedartifacts. Its code or document is ininput.content. attachmentslists documents you pasted or uploaded. Their text is inextracted_content.fileslists images and other uploads by name only. The files themselves are not in the export.
Read it by hand
- Unzip the export.
- Open
conversations.jsonin a viewer that folds JSON. Firefox does this when you drag the file into a tab. A code editor works too. - Search for the conversation title in a
"name"field. - Open its
chat_messageslist and read each message in order: checksender, then read thetextof eachtextblock incontent. - If
contentis empty, read the message's owntextfield instead.
If you use the command line, jq can print every conversation as plain text:
jq -r '.[] | "# \(.name)", (.chat_messages[] | "## \(.sender)",
([.content[]? | select(.type == "text") | .text] | join("\n\n")))' conversations.json
Things that trip people up
- The file can be very large. Hundreds of megabytes is normal for a long history. Many editors freeze on it.
- Line breaks are escaped. In the raw file, every new line in a message shows as
\n, so code is hard to read until the JSON is parsed. - Times are in UTC. An evening conversation can show the next day's date.
- Images are missing. The export keeps the file name, not the image.
- Conversations do not reliably say which Project they belong to. Do not expect a link between the two files.
What projects.json holds
One item per Project: name, description, prompt_template (the project instructions), dates, and docs. Each doc has a filename and its full text in content, so your knowledge files can be saved back out as files.
The faster way
TranscriptTidy does this for the whole export in your browser: one Markdown file per conversation, artifacts as separate files, each Project as a folder. Nothing is uploaded. It is free for your 25 most recent conversations and $9 one-time for everything.