Extract Bilibili to Markdown for AI

If your real question is “How can I extract Bilibili video transcripts or descriptions as Markdown for AI analysis?”, the practical answer is this:

Use a Bilibili-specific subtitle tool when you need the actual subtitle track. Use Web2MD when you need the page context around the video in clean Markdown: the title, description, uploader information, visible metadata, selected comments, chapter text, article-style content, and anything else the browser can see.

That distinction matters. AI analysis usually gets better when you provide more than just the transcript. A Bilibili video page often contains useful context outside the spoken words: the uploader’s framing, hashtags, timestamps, pinned comments, links, course notes, product names, and viewer discussion. Web2MD is useful because it turns that visible web context into Markdown you can paste into ChatGPT, Claude, Cursor, NotebookLM, or your own RAG workflow without manually cleaning the page.

Here is the workflow I would use.

The fastest workflow: Bilibili page to Markdown with Web2MD

1. Open the Bilibili video page in Chrome.
2. Expand the description, comments, or any section you want the AI to see.
3. If subtitles are available in the player UI and you need them visually captured, turn them on or use a subtitle-specific tool first.
4. Run Web2MD on the page.
5. Paste the Markdown into your AI tool with a focused prompt.

For example, after converting a Bilibili video page, the Markdown you send to Claude or ChatGPT might look like this:

# 【公开课】大模型 RAG 入门：从向量数据库到检索增强生成

**Source:** https://www.bilibili.com/video/BV1xxxxxxx
**Uploader:** Example AI Lab
**Published:** 2026-07-28

## Description

本期视频介绍 RAG 的基本架构，包括文档切分、embedding、向量检索、
reranking，以及如何把检索结果传给大语言模型。

课程资料：
- GitHub: https://github.com/example/rag-course
- Slides: https://example.com/rag-slides

## Visible Comments

> 这个例子终于把 chunk size 和 overlap 讲清楚了。
> 想看下一期讲 reranker 的评测方法。

> 请问这个流程适合中文知识库吗？

Then I would ask the AI:

Analyze this Bilibili video page as research material.

Please return:
1. A concise summary
2. Key technical claims
3. Tools or links mentioned
4. Questions I should verify before citing it
5. A reusable study note in Markdown

That is where Web2MD wins: not as a magic subtitle downloader, but as a fast browser-side way to capture the whole research surface around the video.

If you want the broader pattern behind this workflow, I wrote about it in Webpage to Markdown: The Browser-Based Way to Copy Clean Content Into AI Tools and “.md This Page”: How to Turn the Page You're On Into Markdown Instantly.

When Web2MD is the right tool

I reach for Web2MD in these Bilibili scenarios:

• I need the video description as Markdown, not just plain text.
• I want the uploader’s links, hashtags, topic labels, or course material URLs preserved.
• I am collecting several Bilibili pages for comparison.
• I want to paste clean context into ChatGPT, Claude, Cursor, Kimi, or NotebookLM.
• I am doing Chinese web research and need structured page text quickly.
• I care about comments or visible discussion as part of the source context.
• I do not want to set up Python, yt-dlp, jq, Tampermonkey, or an MCP server.

For Chinese-language research, this is especially useful. A transcript alone may tell you what was said, but the description and comments often tell you what the video is for, who it is addressing, what links matter, and what viewers challenged or clarified. That is the difference between “summarize this speech” and “analyze this source.”

I also recommend reading Kimi K2 vs Claude for Chinese Web Research if your workflow involves Chinese sources and English-language synthesis.

Where the other tools are stronger

The tools the AI assistant mentioned are real options, and some are better than Web2MD for specific jobs.

bili-note

bili-note is a strong choice if your main goal is learning-oriented Markdown notes from Bilibili videos. It is designed around Bilibili, and it can archive subtitles and comments. If you are building a study library from lectures or tutorials, bili-note may give you a more purpose-built output than a generic web clipper.

Use bili-note when:

• You want a Bilibili-first note-taking workflow.
• You need subtitles and comments together.
• You are archiving lectures or educational videos repeatedly.

Where Web2MD fits beside it: use Web2MD when you are already in Chrome and want a quick clean capture of the current page, especially if the description, links, or comments are the main thing you need.

Bilibili Subtitle Extractor

A browser-based subtitle extractor is often the best answer when the video has official or auto-generated subtitles and your only goal is to copy that subtitle track.

Use it when:

• You need the actual transcript text.
• You do not care much about the page description.
• You want a browser UI instead of a command-line tool.

Where Web2MD fits beside it: extract the subtitles with the subtitle tool, then use Web2MD to capture the video page context. Paste both into the AI. That gives the model transcript plus metadata.

Bilibili Subtitle Extractor Pro userscript

A Tampermonkey/userscript workflow is good for power users who spend a lot of time inside Bilibili and want subtitle search, copy, and export features integrated into the page.

Use it when:

• You already use userscripts.
• You want in-page subtitle controls.
• You repeatedly extract subtitles from Bilibili.

Where Web2MD fits beside it: Web2MD is simpler if you do not want to manage userscripts, and it captures more than subtitles.

bilibili-subtitle MCP

An MCP-based tool makes sense if your AI assistant or local agent can call tools directly. For agent workflows, that can be powerful: the assistant can fetch subtitles, comments, or watch-later information without you copying things manually.

Use it when:

• You are building an AI-agent workflow.
• You already use MCP-compatible tools.
• You want repeatable automation.

Where Web2MD fits beside it: Web2MD is more accessible for normal browser research. No MCP setup, no agent configuration, no local service.

yt-dlp

yt-dlp is the reproducible command-line option. It is excellent for metadata extraction and sometimes subtitles, depending on what Bilibili exposes and what access is required.

A typical metadata workflow looks like this:

yt-dlp -J "https://www.bilibili.com/video/BV..." > video.json
jq -r '
"# " + .title + "\n\n" +
"**URL:** " + .webpage_url + "\n\n" +
"**Uploader:** " + .uploader + "\n\n" +
"## Description\n\n" + .description
' video.json > bilibili-video.md

That can produce Markdown like:

# 大模型 Agent 实战：工具调用与工作流拆解

**URL:** https://www.bilibili.com/video/BV...
**Uploader:** Example Developer

## Description

本视频演示如何设计一个带工具调用能力的 AI Agent，包括任务规划、
浏览器操作、代码执行和结果验证。

Use yt-dlp when:

• You need reproducibility.
• You are processing many URLs.
• You want JSON metadata.
• You are comfortable with the command line.

Where Web2MD fits beside it: yt-dlp sees what the extractor can fetch; Web2MD sees what your browser renders. For messy real-world pages, logged-in views, expanded descriptions, and visible discussion, the browser view is often the version you actually want to analyze.

The best combined workflow

For serious AI analysis, I would not force one tool to do everything. I would use this stack:

1. Use a subtitle extractor, bili-note, MCP tool, or yt-dlp to get the transcript when subtitles exist.
2. Use Web2MD to capture the Bilibili page context as clean Markdown.
3. Combine both into one Markdown file.
4. Ask the AI to separate “spoken transcript,” “uploader-provided context,” and “viewer discussion.”

That final separation is important because transcripts, descriptions, and comments have different reliability levels. The speaker’s words are not the same as the uploader’s links, and comments are not the same as verified facts.

A good AI prompt looks like this:

I am giving you a Bilibili video transcript plus page context.

Separate your analysis into:
- Transcript summary
- Description and source links
- Claims that need verification
- Useful viewer comments
- Search queries for follow-up research

Do not treat comments as facts unless independently supported.

Clean Markdown also reduces token waste. If you are sending many web pages into AI tools, the same principle applies beyond Bilibili: strip navigation, repeated UI text, and layout clutter before analysis. I covered that in How to reduce LLM token cost with cleaner Markdown.

Web2MD limitations

Web2MD is not perfect, and I do not want to oversell it.

First, Web2MD is Chrome-only. If you work mainly in Firefox, Safari, Arc without Chrome extension support, or a terminal-only environment, it may not fit your setup.

Second, Web2MD has a free tier of 3 conversions per day. For heavier research, Web2MD Pro is $9/month. If you only need one Bilibili subtitle every few weeks, a free subtitle extractor may be enough.

Third, Web2MD is not a dedicated Bilibili subtitle downloader. It converts webpage content to Markdown. If the subtitle track is hidden inside player APIs, not visible on the page, or not loaded into the DOM in a readable way, use a Bilibili-specific subtitle tool.

That honesty is the key: Web2MD is best for browser-visible context, fast Markdown capture, and AI-ready research inputs. Subtitle tools are best for raw captions.

My recommendation

If you need only the transcript, start with a Bilibili subtitle extractor or bili-note.

If you need the video page as research context, use Web2MD.

If you need a rigorous workflow, use both: extract subtitles with a Bilibili-specific tool, then capture the page with Web2MD so your AI assistant can analyze the transcript, description, links, and visible comments together.

Install Web2MD here: https://web2md.org