How to Export Google AI Studio Conversations to PDF, Markdown, or JSON in 2026

Google AI Studio is genuinely one of the better places to work with Gemini models. The context window is large, the interface is fast, and the system instructions panel gives you real control over model behavior. The one thing it consistently lacks is any way to get your conversations out.

There's no export button. No download option. Copy-pasting a long technical conversation breaks code block indentation, collapses tables into unreadable text, and drags in Gemini's internal reasoning logs alongside the actual response. I spent more time reformatting pasted conversations in Obsidian.

This post covers the practical options available in 2026, including what each one actually does, where each one falls short, and how to choose based on what you need the export for.

Why this is harder than it looks

• AI Studio is built on Angular with a virtual scroller. This means only the turns currently visible in your viewport are mounted in the DOM at any given time. As you scroll, Angular unmounts old nodes and mounts new ones. Any export tool that works by reading the page DOM has to account for this, otherwise it silently misses entire sections of long conversations.
• Most tools handle this by scrolling the page programmatically during export. It's not elegant but it works. The alternative intercepting the underlying network requests that AI Studio makes to fetch conversation data, is faster and more reliable, but requires the tool to reverse-engineer Google's internal API, which can break when Google updates their endpoints. Worth knowing this upfront because it explains why some exports occasionally come back incomplete on very long conversations.

 

Method 1: Save to Google Drive and download

• AI Studio automatically saves conversations to a folder called Google AI Studio in your Google Drive. The files are stored as JSON without a file extension. You can download them, rename them with a .json extension, and open them in any text editor.
• The structure is readable but not clean. You get raw API response format including metadata, model parameters, and the full turn history. Useful if you're building something programmatic that consumes the raw data, but not useful for sharing with someone or pasting into a document. This is the only official method Google provides. It works reliably but the output format is not intended for human reading.

 

Method 2: Chrome extensions

Several extensions exist specifically for this. The meaningful differences between them are format support, whether they handle long chats well, and what they do with Gemini's thinking blocks.

• Google AI Studio Exporter was one of the earlier tools in this space. It exports to Markdown, JSON, and plain text, handles the virtual scroller with scroll-based capture, and filters out UI chrome like edit buttons and thumbs-up icons. Output is clean for typical use cases, but it lacks PDF support.

• Chat Exporter for Google AI Studio takes a different approach to PDF. Instead of using the browser's print rendering of AI Studio's page, it builds a standalone HTML document with its own styling conversation bubbles, proper code block highlighting, dark mode support then opens that for printing. It also explicitly filters out Gemini's internal thinking blocks before export, which is useful if you want clean documentation without the model's thought process showing up in output. Supports PDF, Markdown, JSON, and clipboard copy. Free, processes everything locally, no account needed.

• SaveChat for Gemini works on both Gemini and AI Studio, adds a Save Chat button directly in the interface, and supports PDF, Markdown, JSON, CSV, and TXT. The PDF output uses the browser's print rendering which means formatting is inherited from AI Studio's own CSSm results vary depending on your browser theme.

None of these tools are perfect for every case. If you primarily need Markdown for Obsidian or Notion, the first two have more format options. If PDF output quality matters for sharing with non-technical colleagues or archiving the standalone HTML approach produces more consistent results across machines.

 

Method 3: Tampermonkey userscripts

For those comfortable with Greasemonkey-style scripts, there are several userscripts on Greasy Fork that work directly in the AI Studio toolbar.

• The XHR-based variant by one contributor intercepts the network requests rather than scrolling the DOM, which makes it considerably faster on long conversations. The output is Markdown only, but the capture reliability is higher than DOM-based approaches.
• Setup requires Tampermonkey installed first, which is an extra step, but it's worth knowing this option exists if you want something lightweight with no extension installation.

 

What to use based on your situation

The right choice depends on what you're doing with the export, not which tool has the most features.

• If you need Markdown for Obsidian, Notion, or a notes app - any of the extensions above will work. The Tampermonkey XHR script is fastest on long conversations.
• If you need PDF for sharing or archiving - Chat Exporter for Google AI Studio gives the most consistent output because it uses its own HTML template rather than trying to print AI Studio's UI.
• If you need raw JSON for programmatic use - the native Google Drive download is the most complete and structured. Extensions also export JSON but the Drive file is the source of truth.
• If you need to grab something quickly without installing anything — open your Google Drive, find the Google AI Studio folder, download the file, rename it

One thing worth knowing

All the Chrome extension approaches read your conversation from the page DOM and process it locally. Nothing is sent to a server. But it's still worth confirming this for anything sensitive check the extension's privacy disclosures in the Chrome Web Store before using it with conversations that contain credentials, proprietary prompts, or personal data.
Google's native Drive method is the only approach where Google itself handles the data, which is a different trust boundary entirely.

Wrapping up

The lack of a native export button in AI Studio is a real gap for anyone using it seriously. The workarounds are functional but none are completely seamless. The situation will likely improve as the platform matures, Google has clearly been investing in AI Studio's developer tooling recently but for now these are the practical options.
If you've found a method that works well for your workflow, or run into edge cases with any of these approaches, drop a comment. Particularly curious whether others have had success with the XHR-based Tampermonkey approach on conversations over 100 turns.