Large AI apps engage users through intrusive pop-ups.

Author | Wen Changlong Editor | Yang Zhou

A source close to Alibaba's Tongyi App team revealed to "Cityscape" that the Tongyi To C team began emphasizing DAU (daily active users) metrics even before the split from Alibaba Cloud, making it a critical component of interview assessments for certain positions.

This separation was fundamentally driven by the need to bolster DAU. "As early as the beginning of 2024, there were voices questioning why a more capable product team wasn't leading Tongyi App, which was progressing slowly with significant data gaps." Objectively speaking, Alibaba Cloud lacked a To C orientation, and the team consistently faced insufficient resources for App development.

The pressure to increase DAU faced by large tech AI applications is not unique to Tongyi; the only variation lies in how it manifests.

For instance, Baidu's Li Yanhong has publicly advocated for the AI industry to transcend mobile-era frameworks, avoid the "super app trap," and emphasized that success should not be solely measured by one billion DAU. In this context, Wen Xiaoyan's DAU performance might seem adequate, but the internal team's pressure is far greater than it appears.

According to the latest "AI Product Ranking · Domestic Ranking," among apps with tens of millions of monthly active users, Doubao ranks first, Kimi second, and Wen Xiaoyan third. Wen Xiaoyan previously held the second spot, but in December 2024, Kimi surpassed it with 16.69 million monthly active users compared to Wen Xiaoyan's 13.47 million.

"Baidu employees are dissatisfied with Wen Xiaoyan being outpaced by competitors like Doubao, generally believing they 'got up early but arrived late.' However, they understand this is primarily due to Doubao's heavy ad flow investment. Some have questioned why promotional efforts couldn't be increased, to which a leader responded that large-scale Wenxin promotion shouldn't precede profitability."

As everyone knows, user acquisition is crucial for mobile app operations. Acquisition methods are primarily divided into external acquisition and internal mining. External acquisition often relies on resource investments such as ad placement.

Currently, for AI To C apps, ad flow investment has become a key strategy for user acquisition and DAU growth. Reports indicate that ByteDance's Doubao and startups like Kimi have increased their presence on platforms like Bilibili, Xiaohongshu, and Douyin, striving to attract more users.

However, ad flow investment faces two major challenges. Firstly, placement costs are exceptionally high. "Intelligent Emergence" reported that the CPA (cost per acquisition) for Dark Side of the Moon on Bilibili is as high as 30 yuan, and it's rumored that the total ad placement expenditure for AI apps like Kimi, Doubao, and Yuanbao across the network in October 2024 exceeded 300 million yuan.

Secondly, despite significant capital investment, key metrics like product usage duration haven't significantly improved from ad flow investment. According to QuestMobile's third-party monitoring data, over the past year, the average daily usage frequency of apps like Doubao, Kimi, and Wen Xiaoyan has hovered between 4 and 5 times, with an average usage duration of 5 to 10 minutes, showing no significant growth.

Under these circumstances, even big tech companies are becoming more cautious with their spending. When external ad flow investment fails to attract sufficient new users, the focus naturally shifts to the internal market to maintain or increase app activity metrics.

It's evident that big tech AI products like Tongyi App are prioritizing the internal market, offering significant incentives to boost DAU.

01 Is the Direction of Conversational AI Apps Shifting Towards "Newsification"?

When engaging existing users, APP operators typically adopt various strategies to increase DAU. This includes optimizing existing functions (like adding new interaction methods to social apps) or providing personalized content recommendations (like news apps pushing tailored news based on user interests).

Currently, some AI apps seem to be leaning more towards news apps, aiming to encourage frequent app openings among existing users. While this strategy might boost short-term activity, its long-term sustainability is questionable.

Tongyi and Wen Xiaoyan are typical examples. Upon opening the Tongyi App, users are greeted by "Daily News" and "Date and Weather" widgets. Through "Daily News," users can browse trending events, and clicking on them prompts Tongyi to summarize related news and present it in a conversational format.

Moreover, Tongyi actively pushes hot news, triggering a conversation upon clicking to display related news. This push mode is common among major AI apps, with Tongyi standing out due to its frequent pushes.

Tongyi's "Weather" function appears simple but holds a certain appeal. According to a Baidu employee, the daily active users of the weather function in the Baidu App exceed 2 million, with many people opening the app solely to check the weather.

However, this product design has sparked controversy. Recently, user posts complaining about the Tongyi App on Xiaohongshu have garnered widespread attention, with criticism primarily focusing on the news and weather components' design.

A user joked in the comments: "Boss: Our second-half business strategy is to compete with low-frequency apps by being high-frequency, fostering a mindset where users engage with Qianwen daily. Operation: We need to provide more reasons to open it and meet users' diverse needs. Product: Weather and news are live, and the next iteration will integrate Ele.me." Another comment bluntly stated: "Is this still a large model? It looks like a trash can. Which large model app looks like this?"

An Alibaba employee pointed out: "From the App's layout and user experience, Tongyi clearly exhibits a strong traditional traffic operation focus, overly concerned with attracting users' attention while neglecting AI technology's deep interaction and actual product empowerment."

According to "Cityscape," after Tongyi was adjusted to be part of Alibaba's Intelligent Information Business Group, the Tongyi App team is on par with the intelligent search product "Quark," with both To C product design teams being "reused." It's expected that Tongyi App's product design will improve post-adjustment.

Similar to Tongyi, Wen Xiaoyan and Yuanbao's product designs also expose the same issues. Wen Xiaoyan even integrates a news function within the dialog box, where clicking on keywords generates news content. Additionally, Wen Xiaoyan actively pops up news hotspots. A Baidu MEG employee complained to "Cityscape" about Wen Xiaoyan App's chaotic functionality and visual design, bluntly stating that "it wants everything but lacks logic."

Tencent's Yuanbao's initial version also included news hotspots and date widgets on the homepage. Subsequent versions optimized the interface and adjusted functions, gradually consolidating features and changing the news hotspots section's display position, retaining only the news trending chart on the AI search page. Yuanbao also actively pops up news hotspots.

From these details, it's evident that Tongyi, Wen Xiaoyan, and Yuanbao share similarities in design, all attempting to boost DAU by incorporating news and weather components. However, this traditional "traffic-driven" strategy hasn't truly addressed the core issue between AI technology and user interaction experience.

Doubao and Kimi haven't invested heavily in this area. This might be because both products have strong financial backing, allowing them not to be overly distracted by additional features at this stage.

02 Are Big Tech AI Apps Embarking on an Elimination Battle?

At the Volcano Engine Conference on December 18th of this year, "Cityscape" noticed that the Volcano Conference's opening speech was delivered by the Volcano head, focusing on Doubao's visual understanding large model. The second speech was by Zhang Nan, explaining Jimeng, followed by a third speech on the Doubao App.

It's worth mentioning that this was Zhang Nan's first public appearance since resigning as CEO of Douyin Group in February last year to take up the Jianying head position. These arrangements clearly indicate Jimeng's prominent strategic role within ByteDance.

This inevitably brings to mind some background information. Just a week before the Volcano Conference, "Intelligent Emergence" reported ByteDance's internal assessment of AI conversation products' ceiling, believing it might not be high and thus deciding to increase Jianying Jimeng's priority.

The report noted that ByteDance management believes AI conversation products (or chatbot-type products) might only be a "transitional form" of AI products. In the long run, a more ideal product form will likely require a more visual user experience and a lower usage threshold. Therefore, ByteDance has elevated Jimeng's product priority, aiming to create an "AI-era TikTok" through a new path.

Previously, a "LatePost" report also mentioned that some ByteDance employees referred to Doubao as an "ICU product" – it cannot survive without the "intubation" of placement.

Considering this information, while budget-rich Doubao and Jimeng are still competing for the "crown prince" position in ByteDance's AI field, the situation for AI apps from other big tech companies is undoubtedly more complex and challenging.

Take Tencent's Yuanbao as an example. Insiders revealed that the product team primarily consists of employees from other Tencent product lines, without poaching from competitors. Some Tencent employees believe that "Yuanbao currently relies solely on capital investment for traffic acquisition, similar to 'Weishi 2,' but fortunately, Tencent's investment in it is limited, making discussions about Yuanbao somewhat irrelevant."

Internal voices also point out that "Yuanbao differs from Weishi. Weishi's target market was very clear, with huge investments but a complete failure, leaving almost no legacy. In contrast, Yuanbao's target market is inherently volatile, with less personnel investment, and Yuanbao clearly has an advantage in technological legacy."

Furthermore, according to "Cityscape," a dialog-based AI app under the BAT series had a mere four-digit daily active user count for a period, despite frantically engaging in public relations and promotion. Internal employees complained that this product was merely a promotional tool for senior management.

Generally speaking, in this AI wave, the market is stratified into three opportunities: model providers, cloud vendors, and AI product applications. In this ecosystem, big tech companies, with their accumulated advantages in capital, talent, and resources, can often dominate these markets.

Currently, general large models have evolved into a "super money-burning" game, becoming a "rich people's game" exclusive to giants. Companies lacking sufficient funds will have to withdraw from this competition.

The previous consensus was that to excel in the "money war" of foundational large models, besides having short-term financial support, long-term self-sustaining capability is crucial, with AI applications being the key direction to achieve this.

However, currently, big tech companies' first attempts at AI applications seem to have fallen short of upholding this banner.

Entering 2025, in the large model field, whether it's the base model layer or the AI application layer, competition among big tech companies will intensify. Against this backdrop, will any big tech company's AI application be the first to be eliminated?