Reference
File formats: what each export contains
Backup My AI Chat writes ordinary files: PDF and Word for people, Markdown and text for notes and editors, JSON for scripts, and images for sharing. This page lists what each one holds, how auto-saved files are named and arranged, and the settings that shape them, so you can decide which format to use and build on the files with other tools.
The formats at a glance
| Format | Extension | Best for | Where you get it |
|---|---|---|---|
| Sending, printing, keeping a fixed copy | Export, Custom Export, Bulk Export | ||
| Word | .docx | Editing, Google Docs, handing to a client | Export, Custom Export, Bulk Export |
| Markdown | .md | Notes, Obsidian, Git, search | Export, Custom Export, Bulk Export, Auto-save, side panel |
| Plain text | .txt | The words and nothing else | Export, Custom Export, Bulk Export |
| JSON | .json | Scripts and other software | Export, Custom Export, Bulk Export |
| Image | .png | Posting a chat where files are awkward | Export, Custom Export, Bulk Export |
| Image archive | .zip of .png | One image per message | The More menu, Custom Export |
| Shareable image | .png | A single styled image for sharing | The More menu, Custom Export |
Markdown and the file header
Markdown is the format Auto-save uses, and the one to choose for anything you will search, edit or keep for years, because every text editor opens it. Each file starts with a header between two --- lines that lists title, source, conversation id, link, created, updated, model, message count and tags. Obsidian shows the header as properties; other tools read it as YAML.
---
title: "Sorting a list in Python"
source: chatgpt
id: "…"
url: https://chatgpt.com/c/…
created: 2026-03-04T09:12:00Z
updated: 2026-03-04T09:20:41Z
model: "gpt-5"
messages: 2
tags:
- ai-chat
- chatgpt
---
# Sorting a list in Python
## You · 2026-03-04 09:12
How do I sort a list of dictionaries by a key?After the header comes the conversation, with one heading per message that names the author and the time. Code blocks stay as fenced code with their language, tables stay as tables, and maths is written in dollar-sign notation. One-off Markdown and text exports can also start with the chat's source URL, a setting under General.
The Auto-save folder
Auto-save creates one subfolder per service inside the folder you chose, and project chats get their own folders. Files are named with the date and the chat title; two chats with the same date and title get distinct names.
AI chats/
├── ChatGPT/
│ ├── 2026-09-12 Chat number 8.md
│ ├── 2026-09-13 Chat number 7.md
│ └── Projects/
│ └── Research 2026/
│ └── 2026-09-05 Project chat.md
├── Claude/
│ ├── 2026-09-16 Claude chat 5.md
│ └── Projects/
│ └── Website/
│ ├── Project instructions.md
│ └── Knowledge/
│ └── glossary.md
└── Grok/
└── 2026-09-14 Grok talk 1.mdA chat that changes updates its own file, a renamed chat keeps its file, and a file you have edited is left alone unless you switch Overwrite files I have edited on.
Attachments and images
Images and files uploaded in a chat are saved, where the AI site allows it, in a _files folder next to the chat and linked from the Markdown. For ChatGPT and Grok, some images live on a separate file server; the Auto-save page asks once for permission to read it with an Allow access button, and without that permission those images are noted in the text instead.
Service by service, Auto-save also keeps: ChatGPT archived chats, project chats and Canvas documents (first and final version); Claude project chats, project instructions, project knowledge files, artifacts (first and final version) and the text of uploaded documents; Grok generated images; Perplexity sources as a numbered list; DeepSeek thinking, citations, sources and file names; Gemini thinking and image notes.
JSON
The JSON export is structured data for programs: the chat title, the platform, the model, the URL, and the messages, each with a role and its text. Use it to feed a conversation into a script, a database or another tool. For reading, choose Markdown instead.
PDF and Word
PDF and Word exports are rendered with the Style Settings: one of eight themes (Light, Dark, Note, Old Paper, Sakura, Blue Mist, Fresh Green, Lavender), a page width, a font size (small, medium or large), and whether each message shows its time. Custom Export lets you change theme and size for one export without changing the defaults. Word files open in Microsoft Word, Google Docs and LibreOffice.
Plain text
The text export holds the conversation as plain text with the author labels and, if you want, the message times and the source URL at the top. It is the smallest file and the one that pastes cleanly anywhere.
Image, image archive and shareable image
Image is one PNG of the whole chat, rendered with the Style Settings. Image archive is a .zip with one PNG per message, for picking out single answers. Shareable image is a single styled image meant for posting. In Bulk Export, several chats become one long image or a section per chat.
Filenames
One-off exports are named after the conversation title by default. Under Settings → General you can choose date plus title, or a pattern built from {title}, {date}, {time}, {platform} and {model}. Bulk Export uses the same pattern. Auto-saved files are always named with the date and title.
What every format keeps
- Headings, lists, tables, code blocks with their language, inline code, links, bold and italic, and maths.
- The model's thinking section where the site shows one, if the thinking setting is on.
- Author labels (You, or the assistant's name), the model name, and optionally the message time.