Voice Input in China Is Learning to Write for You: From Dictation to Finished Text

Within a year, the market moved from a first launch to four major players. Each one is trying to claim the spot through which almost everything a person writes passes.

• November 2025: launch of Doubao Input • December 2025: desktop AI input from Zhipu • May 2026: Qianwen voice input; WeChat has its own feature, "隔空传送"

A tool that "already satisfied everyone" has turned, within a year, into one of the tightest arenas in Chinese AI. Between November 2025 and May 2026, at least three new AI keyboards from major players hit the market, and WeChat added a feature called "隔空傳送" ("transmission through the air") to its input method. The main front in this battle is now voice input, and the bet being made is a shift in the core task: the system doesn't listen to words — it listens to the user's intent and immediately produces finished text.

Timeline: four moves in seven months

The countdown starts in the fall of 2025. Here's what's known about how events unfolded:

• Doubao Input: launched in November 2025. Doubao is ByteDance's AI assistant, and the keyboard became a new entry point for it on phones.
• Zhipu's desktop AI input: December 2025. Zhipu is known as a developer of its own large models and went not for mobile but for the desktop segment.
• Qianwen voice input: May 2026. Qianwen is the consumer AI brand of Alibaba's ecosystem, and this time the emphasis is specifically on voice.
• WeChat input with the "隔空傳送" feature: a tool for quickly transferring typed text between devices. It leans on the messenger's enormous user base.

The launches come in rapid succession, roughly every one to two months. For a market that used to change over the course of years, this is a very fast pace.

Why the keyboard became a strategic beachhead

The keyboard sits at the start of almost every user scenario: a message, an email, a search query, a note, a prompt for a chatbot. Whoever controls this layer sees the user's intent before any app does and can embed a model directly into the moment text is created. For a company with its own LLM, this is the shortest path from model to daily habit: there's no need to ask the person to open a separate chat.

Voice amplifies the effect. Ordinary dictation has long been the norm, but it requires editing: people speak with pauses, repetitions, slips of the tongue, and jump between thoughts. The new generation of AI input methods takes on that work. The user states the gist — for example, "tell my colleague the meeting's moved to Thursday, and apologize" — and gets back a neat, appropriately toned message. Transcription turns into generation, which means quality now depends not only on speech recognition but on a language model that understands context and style.

What this changes for the market

This approach has weak points too. Generation instead of transcription means the system can rephrase what was said in a way the speaker didn't intend. That raises a trust question: how accurately does the final text convey the thought, and where's the line between help and substitution? The second problem concerns data. The input method sees everything a person writes and says, so questions about on-device versus cloud processing become just as important for keyboards as recognition accuracy.

At the same time, the competition isn't playing out along a single axis. Doubao and Qianwen are tied to their AI assistant ecosystems, Zhipu chose desktop, and WeChat is leaning on cross-device handoff plus its built-in user base. The likely winner will be whoever combines model quality, response speed, and convenience in everyday use most effectively.

What to look for when choosing

If you're trying out these input methods, check three things. First: how the system handles speech that mixes languages and technical terminology. Second: whether you can quickly see the original transcription alongside the finished text, to catch distortions. Third: where the voice data is processed and what gets stored. For teams and companies building products on top of LLMs, the takeaway is simpler: the field where a model meets the user at the exact moment text is created has turned out to be no less valuable than the chat window — and competition there is just getting started.