How to Save AI Chats to Obsidian

Quick answer: To save an AI chat to Obsidian, export the conversation as a Markdown file and drop it into a folder inside your vault; Obsidian reads it immediately, with headings, tables, code blocks and LaTeX maths rendered.

The Claude and Gemini extensions have a Markdown button for a response or a whole chat. Add front matter with the assistant, date and tags, and link the note from the project it belongs to.

Why a vault suits chat archives

A conversation with an assistant is a working document that often contains the one paragraph you'll want in six months.

Left in the assistant, it's one of hundreds of untitled chats, searchable only through the assistant's own search, and subject to deletion (Gemini auto-deletes by default).

In a vault it's a text file:

• searchable
• taggable
• linkable from the notes it fed
• backed up with everything else
• readable by anything that reads text

Getting a Markdown file of the chat

• Claude: enable the Markdown button in the Claude to PDF, Word and Google Docs extension's settings and click it at the top of the chat or under one answer.
  • The file is made in your browser and embeds images as data URIs, which Obsidian renders.
  • For one answer, Claude's copy button already copies Markdown. Claude to Markdown.
• Gemini: the Gemini extension's Markdown button exports a response or the whole conversation, loading earlier messages first. Gemini to Markdown.
• ChatGPT: no Markdown export in the extensions.
  • ChatGPT's copy button copies an answer as Markdown; paste it into a new note.
  • For whole conversations, export Word (ChatGPT to Word) and convert with Pandoc (pandoc chat.docx -t gfm -o chat.md) or via Google Docs' File → Download → Markdown.

Where the files go

A structure that scales:

AI/
  ChatGPT/
  Claude/
  Gemini/
Projects/
  project-name/

Chats live in the assistant's folder under AI/, named YYYY-MM-DD topic.md so they sort by date. Project notes live under Projects/ and link to the chats they used.

Keeping chats out of project folders keeps those folders readable; the links carry the relationship.

Front matter that pays off

At the top of each imported chat:

---
assistant: claude
date: 2026-10-17
tags: [ai-chat, pricing]
source: https://claude.ai/chat/…
status: reference
---

With properties like these, Obsidian's search and the Dataview plugin can answer "every Gemini chat about pricing since September" or "chats marked final".

The source link is the original chat while it still exists; it's not a backup, the file is.

What renders, and what to fix

• Headings, lists, bold, links: as written.
• Tables: pipe tables render as tables. Very wide tables scroll horizontally; that's a display choice, not a loss.
• Code blocks: fenced blocks render with syntax highlighting when the language is given.
• Equations: LaTeX between $…$ and $$…$$ renders through Obsidian's built-in MathJax (Obsidian's formatting help), so Claude's and Gemini's maths display correctly without changes.
• Images: data-URI images (Claude's export) render inline.
  • URL images render only while the URL works.
  • For anything you care about, download the image into an attachments folder and change the reference to ![[image.png]].
• Prompts: the exports mark your prompts as blocks or quotes. Obsidian renders quotes as callout-like blocks; if you'd rather have Q and A headings, a quick find-and-replace does it.

Queries that make the archive useful

Once chats carry front matter, the Dataview community plugin turns the folder into a database. Three queries cover most needs:

• Everything from one assistant this month: a table of file.link and date where assistant = "gemini" and date >= date(2026-10-01), sorted by date. Useful when you suspect you asked the same question twice.
• Chats behind a project: a list of notes under AI/ that link to the current project note, which is the reading list for anyone picking the project up.
• Reference notes not yet summarised: notes tagged ai-chat with status = "reference" and no summary property. That's your to-do list for turning raw transcripts into notes worth keeping.

Without Dataview, the same questions are answered by Obsidian's search with property filters ([assistant:gemini]) and tag searches; the plugin only saves retyping them.

A routine that keeps it manageable

Archive on a schedule rather than chat by chat. Once a week:

1. Export the conversations from the past seven days that you'd mind losing.
2. Drop them into the vault.
3. Spend ten minutes adding a two-line summary at the top of each.

A transcript with a summary is a note; a transcript without one is a file you'll never open.

Delete the chats you exported from the assistant if you prefer a clean history there; the vault copy is the one that lasts.

Linking chats to the work they informed

The point of the archive is the links.

When a chat produced a decision, a draft or a number, add [[2026-10-17 pricing options]] to the project note where that decision lives, and a line under it saying what was taken from the chat.

The backlinks pane on the chat note then shows everywhere it was used, which is the audit trail an assistant can't give you.

For a hosted, collaborative version of the same archive, see AI chats to Notion; the Markdown files are identical, only the destination differs.