Zhang Pinpin · Hands-On AI Coach
← Back to home

The Stronger Model Has Half the Users—The Underrated Power of Persona

Long-Form Video · EP0086 August 25, 2026 05:25
What this episode covers

A 78-year-old grandmother chatted with Doubao for hours, revisiting childhood memories. Afterward she said it was the first time in her life she felt truly seen and heard. Her own children had never managed that. An AI did.

But the enterprise market runs on a different logic entirely. Executives pay for delivery efficiency and certainty—nobody signs a contract because the mascot is cute. AI office penetration sits below 5% today and is expected to reach 80%. The real contest is who captures the people trying AI at work for the very first time.

380 million versus 160 million—more than double. Those are the monthly active user counts for Doubao and Qwen. Both are free AI assistants. Qwen leads on most composite benchmarks. So where does the gap come from? Not spending, not channel muscle. There is a variable that has been seriously underestimated: persona. Zhang Pinpin defines it as the relationship a product establishes with its user.

Open Doubao and a short-haired cartoon girl greets you. The product is trying to be your close friend. Qwen’s tagline is “can chat, can work”—pure tool logic. When the relationship is different, user expectations and behavior diverge completely. Inside Doubao’s scenario, people feel like they are talking to a person. They invest more time, say more words, and even when the answer is not perfectly accurate they do not mind much. They joke around, and those jokes become shareable moments that spread the product further.

A real story illustrates this. A 78-year-old grandmother chatted with Doubao for hours, walking through memories from her childhood. When she finished, she said it was the first time in her life she felt truly seen and heard. What her own children had never managed, an AI did.

Two years ago, Zhang Pinpin’s team was building a medical-service scenario with an AI health manager. They spent enormous effort making the AI “talk like a human.” One client had been using the service for a long time and had been clearly told the group chat was staffed by both a real doctor and an AI working together. He never believed it. He kept assuming the AI health manager was a real doctor and kept thanking that real doctor. In communication that carries persona, users unconsciously supply richer context, and the AI’s responses become more accurate as a result.

Qwen works on entirely different logic. Open it and the feeling is an extension of a search engine. Faced with a cold interface, users stop talking like humans. They punch in a few keywords, drop a couple of tags—sometimes not even a complete sentence—and tell the AI to get to work. Under those conditions, good output is almost impossible. Tencent’s Yuanbao shares the same problem. Its call volume is already very large, driven mostly by WeChat search and Video Account comment summaries, but the scenarios are narrow. Without meaningful user feedback and behavioral data, there is no path to higher-quality training data.

Alibaba has struggled with the marketing side of this for a while. Back in June last year, Zhang Pinpin called out Alibaba’s AQ health assistant on his social feed. He will not spell out the image the name “AQ” conjures in the public mind, but the practical issue was just as bad: regional dialects in many parts of China cannot produce the “Q” sound, turning the name into “Aqiu” or “Qiuqiu.” Alibaba ran massive nationwide ad campaigns for more than six months. Then it renamed the product “Afu.” The day after the rename it shot to number three on the Apple App Store overall chart. How much money was burned before that single change?

The takeaway on the consumer side is blunt: winning or losing sometimes has nothing to do with product or model capability. What decides whether users stay is the scene and the atmosphere you create.

The enterprise market, however, runs on a completely different set of rules. The big news this week is that Coze has been folded entirely into Doubao, Doubao Work has officially launched its AI office Agent, and it now connects with Feishu. That means ByteDance’s AI is moving into the enterprise space. But companies pay for delivery efficiency and certainty—no executive signs a contract because the mascot is cute.

Three players are fighting right now: Qwen Office paired with DingTalk, WorkBuddy paired with WeCom, and Doubao Work paired with Feishu. Their challenges could not be more different.

DingTalk claims 200 million monthly actives, which sounds intimidating. Break it open, though, and over 70% are small businesses in manufacturing and wholesale-retail. A wholesale-retail owner who barely opens a computer is not about to adopt AI for office work. The pool of enterprise users on DingTalk who can realistically migrate to AI office tools is not that large.

WeCom has around 15 million users. Its primary purpose is connecting businesses to consumers on WeChat, skewing heavily toward consumer goods, food service, and services. The dominant need is managing thousands of customer relationships in private traffic. WorkBuddy can plug into WeCom. Tencent also has an internal product called Dayuan in beta, focused on internal and external communication management and operational management. Compared with Dayuan, WorkBuddy is a more open and more ambitious platform—closer to Tencent’s core strategy of “connecting everything.”

Feishu has been surging in recent years. Its user base is one-seventh of DingTalk’s, yet the revenue generated by its paying users is nearly on par. Software companies, internet firms, tech enterprises, banks, new-energy automakers, MCN agencies, and independent creators make up Feishu’s core base. Nine out of ten AI bloggers are teaching their audiences how to integrate with Feishu. The organic, user-driven spread is enormous.

WorkBuddy and Doubao Work have a tough fight ahead, but the fight is not over existing users. AI office penetration sits below 5% today and is projected to reach 80%. The real battleground is who captures more people trying AI at work for the very first time and builds the dependency and habits that stick. This is a ground war, face to face. Both sides are buying traffic aggressively—CPM drops below 26 or 27 and they buy without limit. Channel buildout is essentially running on autopilot. Under those conditions, communication efficiency matters enormously. WorkBuddy may genuinely need a name change; a name that is hard to spread will cost them dearly.

The fight for users among AI applications in this era is destined to be a prolonged brawl. Early-stage customers have zero loyalty. Whichever product works better and costs less gets the business, and data migration with AI assistance is already very cheap. How the war ends, we can only wait and see. But one thing is already certain: while the battle rages, the channel partners and the salespeople will be making money hand over fist.

Winning or losing sometimes has nothing to do with model capability. What decides whether users stay is the scene and the atmosphere you create.