ByteDance Built an AI That 600 Million People Use. Outside China, Almost Nobody Has Heard of It.
Doubao. 豆包. ByteDance's answer to ChatGPT. If you live in China or follow Chinese tech, you already know what this is. If you don't, here's the short version: the same company that built TikTok took their large language model and wrapped it in a chatbot interface that now has more monthly active users than Claude, Gemini, and Perplexity combined. It's absurdly popular in China. Students use it for homework. Writers use it for brainstorming. Developers use it for code review. My cousin in Beijing uses it to plan her weekly grocery shopping and honestly the meal plans it generates are better than mine.
I started using Doubao three months ago when I was learning Mandarin and wanted an AI that could explain Chinese grammar patterns in Chinese, with examples, without defaulting to English. ChatGPT kept switching to English mid-explanation. Doubao stayed in Chinese. And it explained things the way a patient tutor would — not just the rule, but the exceptions, the common mistakes, the situations where native speakers break the rule on purpose for emphasis. The explanations averaged about 1,200 words. Beautiful, clear, completely useless to me in their written form because I was reading Chinese on a screen at a pace that made my eyes cross after paragraph four.
I needed to hear it. Not read it. Hear the words spoken naturally so my brain could connect the written characters to their pronunciation while absorbing the grammar explanation. Reading Chinese is slow for me. Listening to Chinese is how I actually learn.
CastReader solved this. It's a free Chrome extension. Install it, open Doubao, and each AI response gets a speaker button. Click it. The response reads aloud in natural voice with each paragraph highlighting on the page as it goes. For Chinese content, this means I hear the pronunciation while seeing the characters light up on screen — essentially a synchronized reading and listening experience that's closer to having a tutor read to me than anything else I've tried.
The technical side of why this works on Doubao is worth explaining because Doubao's page structure is unusual. Most AI chatbots use standard HTML paragraph tags for their response text. Doubao doesn't. ByteDance's frontend team built a custom rendering system — response text lives inside div elements with dynamically generated class names like "paragraph-pP9ZLC" and "paragraph-Qn8mXK". These class names change. There are no stable data attributes to grab onto. The action buttons (copy, regenerate, share) float in containers with similarly opaque class names. A generic text extraction tool would either grab everything including the UI elements or fail entirely because it's looking for standard HTML patterns that don't exist on Doubao's page.
CastReader has a specialized Doubao extractor that handles all of this. It identifies AI responses through the ".flow-markdown-body" containers, matches paragraph elements using class-name pattern matching, filters out code blocks and action bars, and extracts just the response text in the correct reading order. I didn't have to configure anything. I installed CastReader, opened Doubao, and it worked. The extractor detects the site and engages automatically.
There's a specific use case where Doubao plus CastReader is genuinely better than any alternative I've found: bilingual content. I frequently ask Doubao questions in English and get responses that mix Chinese and English — a Chinese explanation with English technical terms, or a Chinese-English comparison with examples in both languages. Reading bilingual content on screen is cognitively expensive because your brain has to switch parsing modes for every language boundary. Listening to it is surprisingly smooth. The voice handles the language transitions naturally. A sentence in Chinese flows into an English term and back to Chinese without awkward pauses or pronunciation disasters. My Mandarin tutor heard it during one of our sessions and asked me what app I was using. I showed her CastReader and she downloaded it before the lesson was over.
I also use Doubao for creative writing in Chinese — short fiction, dialogue practice, poetry analysis. Doubao is remarkably good at this, probably because ByteDance has access to an enormous Chinese text corpus. But creative writing responses are long. A character analysis of a Tang dynasty poem can run 1,500 words. A short story continuation is easily 2,000. Reading this on screen turns art appreciation into homework. Listening to it turns it back into something closer to storytime. The voice gives the text a rhythm that my eyes can't.
My friend in Shanghai — the one who first told me about Doubao — uses CastReader to listen to Doubao's coding explanations while he's debugging. He asks Doubao to explain a section of code, clicks the speaker button, and listens while he works in his IDE on the other monitor. "Reading the explanation and the code at the same time is impossible," he told me over WeChat. "Listening to the explanation while reading the code is easy." He said it in Chinese and it sounded more elegant than my translation, but the point stands.
The setup is trivial. Chrome Web Store. Install CastReader. Free. Not "free trial," not "free for 500 characters," not "free with registration." Free. No account, no email, no credit card. Open www.doubao.com, start a conversation, wait for a response, click the speaker icon. A floating player appears at the bottom of the page. Pause, resume, speed control, paragraph navigation.
CastReader also works on DeepSeek, Kimi, ChatGPT, Gemini, and Claude. Each platform has its own specialized extractor. But Doubao is where CastReader felt most necessary to me — because Doubao is where the content is in Chinese, and Chinese is where my reading speed is slowest and my need for audio is highest. Six hundred million users generate a lot of knowledge in those conversations. Being able to listen to it instead of reading it changed Doubao from a tool I admired into a tool I actually use.
