AI application is missing a Zhang Xiaolong

New stance

follow with interest

2025-03-15

0 comments

298 views

1 Collection

14 minutes

  Product managers need to focus on different aspects at different career stages, ranging from basic skills, business depth, professional fields to strategic planning and management abilities.

From Doubao to Manus, and then to DeepSeek, various AI applications have shown their unique abilities in the race to compete for user minds. However, can these popular applications really meet the needs of consumers? Do they represent the ultimate form of AI consumer applications? This article will analyze the current status and challenges of AI applications from multiple dimensions such as technological inclusiveness, user experience, and market promotion, and explore possible future development directions.

It has been over two years since ChatGPT ignited the AI topic, but there has not been a dominant AI application in China. It can even be said that until the beginning of this year, a series of discussions around DeepSeek radiated the whole Chinese Internet, consumer level AI applications really had a chance to be on the table.

The reasons for this are multifaceted. The cognitive leap from cutting-edge technology to the mass market requires both time and a product that knows how to bridge both ends and design outstanding products.

But at the beginning of the year, a domestically produced large model brought AI to the public's attention, and the existence of the technology itself gained public recognition. In addition, the question of "what kind of AI applications do consumers really need" has gradually emerged through the exploration of numerous products and feedback from a large number of users. Including the recently discussed Manus, they are catalyzing the formation of a consensus on AI applications in the entire market.

Renowned consulting firm a16z released its latest global AI application ranking at the beginning of this month, based on Sensor Tower's monthly active user data. Only 6 months have passed since the last ranking update, and we can already see a major reshuffle in the landscape. DeepSeek, which has surpassed 10 million users within 20 days of its release, has also made the list, while domestic applications such as Doubao, Baidu, and Meitu have experienced changes in their rankings.

Global AI Application Ranking (2024.8/2025.3); Source: a16z

At this point in time, the development threshold for AI applications has lowered, and the freedom to customize products based on consumer demand has increased. The commercialization efficiency that was temporarily put on hold in the past is also gaining weight.

In the statistics of a16z, 40% of high-income applications did not enter the top 50 users. The most popular applications among consumers may not necessarily have advanced or enterprise level features that professional users are willing to pay for.

When AI truly begins to show signs of technological inclusiveness, transforming materials into finished products, it is the moment to test and teach all parties to "forge" technology. At present, there are indeed some cases worth referring to.

Just like instant messaging technology has long been mature, but it was not until the birth of WeChat that it truly "unified the world". There is no end to breakthroughs and refined research in underlying technology, but now it is time for product managers to make efforts.

The universality of high-dimensional technology is based on the premise of "sinking"

Cognition is an essential prerequisite for consumption, and the same applies to the application of consumer goods.

The product oriented marketing concept believes that as long as the technical skills are in place and the product is excellent enough, it can attract consumers through word-of-mouth. But modern marketing is not the era of 'the fragrance of wine is not afraid of deep alleys', users receive hundreds or thousands of advertising messages every day. A report in 2022 pointed out that the continuous focus time of contemporary people has decreased from 12 seconds in 2000 to 8 seconds today, comparable to goldfish.

AI applications are more unique due to the presence of dual cognitive barriers. Firstly, in terms of technical understanding, the vast majority of users have no concept of the operational logic of deep learning and AIGC. In addition, the services provided by AI have a certain perceptual lag. Unlike communication tools such as WeChat, which address explicit needs for instant communication, AI does not directly provide sensory stimulation as entertainment applications. Without practical experience transfer, the market education difficulty of AI applications is greater, and promotion relies heavily on hot start.

How can B2B product managers grow rapidly?

Product and business architecture mainly involves layering and organizing the entire business workflow, abstracting individual requirements, mapping business requirements to products in a reasonable and reasonable manner, and ultimately enabling business data to flow, execute, record, and use within the product.

View details>

Therefore, relying on Tiktok and WeChat, the promotion path of Doubao and Yuanbao with original flow pool is the most typical.

Dou Bao has made the "sinking" more thorough, and its "uniqueness" can be distinguished at first glance. There are three schools of AI application naming, or ERNIE Bot (Wenxiaoyan), Tongyi Qianwen, or Kimi, DeepSeek, or "Doubao". When a group of AI application icons pursue minimalism and a sense of technology in design, Dou Bao chose an extremely "approachable" and AI flavored cartoon big head portrait.

Backed by ByteDance's powerful promotion channels, Doubao quickly achieved a user scale gap. According to the report of Tianfeng Securities, the amount of doubao launched in April and May last year was nearly 18 million yuan, which rose sharply to 124 million yuan in the first ten days of June, of which 89% of the advertising proportion was invested in the massive engine, and bytes also limited the advertising of Kimi and other competitors on the Tiktok platform.

There is a theory of "digital divide" in communication studies, which refers to the polarization of information gap caused by different media literacy, such as middle-aged and elderly people who are often blocked by the digital divide from new technologies. With the tentacles of short videos in various circles, Doubao may be the AI application with the highest recognition in the sense of "pan sinking". Its strategy is to replace the high-tech facade with cartoon characters, incorporating topics such as "AI parenting" and "Spring Festival blind boxes" into daily agendas. Professionalism is secondary, and users should first have a sense of interaction and gain.

Tencent's response to the AI application side may have been somewhat belated, but Tencent Yuanbao's recent advertising efforts have shown a momentum of being a latecomer and first mover.

Yuanbao was launched in May last year, and in its early days, it did not have a strong presence among a group of general AI applications. At that time, the investment battle for apps such as Doubao and Kimi had already begun. But after DeepSeek became popular, Tencent went against the defensive and launched a full-scale attack. The strategy was also very straightforward, which was to use the opportunity to promote and write "fully loaded DeepSeek" in the name suffix.

According to AppGrowing data, the estimated investment amount for Yuanbao within 30 days has exceeded 500 million yuan. In addition to large-scale advertising in various channels, Tencent's own ecosystem promotion is almost used to the extreme, including the golden recommendation slot of WeChat life service "Nine Palace Grid".

If Dou Bao had made up his mind to take the "strong push" route from the beginning, Yuanbao would have had the opportunity to overtake in a bend. This is not only due to the current high market sentiment, but also because the intellectual competition of AI applications will no longer give lagging players too much time. Moreover, gaining awareness is only the first step.

02 Easy to use does not equal practicality

QuestMoblie's report at the beginning of the year showed that the stickiness of integrated native apps improved significantly in the second half of last year, but most still face the challenge of high uninstallation rates, and the value of AI has not yet been fully realized. It is also because of the high uninstallation rate that companies need to constantly attract new customers and continue to invest in order to consolidate their market share.

Source: QuestMoblie

Entering the level of product quality evaluation, "ease of use" is probably the first threshold, and its most intuitive evaluation criterion is to avoid middlemen making price differences as much as possible.

The profit from reselling AI "knowledge" is almost an enduring news after the popularity of ChatGPT, but the protagonist has recently been replaced with DeepSeek with version updates. Due to the high technological threshold and strong demand for paid knowledge, a large number of opportunities for "information gap" arbitrage have emerged as a result of catalyzing anxiety through the topic of technological iteration.

In the era of ubiquitous applications, many products have put effort into usability in order to provide differentiated services. For example, providing customized scenarios, well-trained agents, and templates that can be directly applied to avoid the "Prompt Gap". The core logic is to reduce the operational steps and learning costs required for users to achieve their goals.

The significance of ease of use for consumer applications can refer to another case study. PhotoShop has dominated graphic processing tools for over a decade, giving rise to a large number of design courses and thriving outsourcing agencies for over a decade. But Canvas greatly reduces the threshold for use through web entry and template based operations, allowing users to complete designs with limited freedom. This is the most typical difference in purpose between the technical foundation and product design.

But further, whether AI generated content can be delivered smoothly in real application scenarios is still in the stage of chaos among all parties.

The most obvious problem is how to stand out in homogeneous output. At present, C-end AI applications can be roughly divided into two types: native independent apps and plugins (embedded in existing apps), covering scenarios such as education and learning, leisure and entertainment, image generation, life assistants, emotional companionship, and more.

If I were a self media copywriter and wanted to design copy for daily updates on social media platforms, I could find at least 5 AI tools that could achieve this goal, some even providing various scene templates. Furthermore, there are already content platforms that can generate copy with just one click using AI, eliminating the need for outsourcing programs.

The judgment logic of "practicality" is that the system can stably and reliably output results that meet expectations in real scenarios. At present, users spend most of their time using AI applications making this judgment action, and it is obvious that there are no applications that are leading the gap. The underlying support for 'usability' is the thickness of technology, and DeepSeek has become the current focus on this.

However, "New Position" believes that Manus, which has recently gained enough attention, has put forward a new perspective on product design, setting aside marketing controversies.

The thinking process demonstrated by DeepSeek allows users to intuitively see the generation logic of AI, which was a highly praised aspect when it was first mentioned. Manus to some extent amplifies this sense of achievement, as the interactive logic of its product is that it only requires a simple instruction for AI to complete the process from breaking down requirements to calling different APIs to execute tasks step by step.

The full process visualization, coupled with user adaptability, to some extent combines usability and practicality. Although there is still room for improvement in the specific output effect, it has indeed brought some experiential revolutions.

This suggests that as a groundbreaking technology, the design of its practical applications can go further and attempt to redefine and transcend the original paradigm.

03 written at the end

According to QuestMoblie, 70% of apps with a scale of over 100 million are on the path of transformation, either arming their core businesses with AI plugins or extending their scenarios through AI exploration. For example, Zhi Xiaobao, Alipay's life service assistant, Weibo Smart Search, a new interactive sector launched by Weibo, and the "DeepSeek" trend that has swept the entire Internet service in recent two months.

That's also why we say that AI applications have truly entered the "battle" stage. When various tools and entertainment within reach are embedded with AI functions, users' perception, evaluation, and other feedback information about it will grow exponentially, and the differentiation of retention rates will be more obvious.

At present, the vast majority of users are only accustomed to using one comprehensive native app, accounting for over 80%. Only 1% of users are willing to use four or more apps for multi-party comparison, and the Pareto Principle is evident.

In other words, although AI will probably become the "hydropower coal" of the Internet, only a few mature AI products can pick up the sweetest fruits.

*The title and accompanying images in the text are sourced from the internet.

Author: Wang Wei Editor: Li Fan

Source official account: New Position Pro (ID: xinlichang66), looking at the world from a different perspective.

This article is authorized to be published by the collaborative media @ New Position, where everyone is a product manager. Reproduction without permission is prohibited

Image from Pixabay, based on CC0 protocol

The viewpoint of this article only represents the author himself, and everyone is a product manager. The platform only provides information storage space services.

AI 应用缺一个张小龙

新立场关注

2025-03-15

0 评论298 浏览1 收藏14 分钟

从豆包到Manus，再到DeepSeek，各款AI应用在争夺用户心智的赛道上各显神通。然而，这些热门应用真的能满足消费者的需求吗？它们又是否代表了AI消费级应用的最终形态？本文将从技术普惠、用户体验、市场推广等多个维度，剖析当前AI应用的现状与挑战，并探讨未来可能的发展方向。

ChatGPT 点燃 AI 话题已经过去两年多，但国内并没有出现统治级别的 AI 应用。甚至可以说，直到今年初围绕 DeepSeek 产生了一系列辐射整个中文互联网的讨论之后，消费级别的 AI 应用才真正有了上桌的机会。

其中的原因是多方位的。从前沿技术到大众市场的认知跨越既需要时间，也需要一款懂得弥合两端、设计出彩的产品。

但年初一款国产大模型将 AI 带到全民关注，技术本身的存在有了大众认知度。再加上，“消费者到底需要什么样的 AI 应用”这个问题，在众多产品的摸索、大量用户的反馈中，逐渐有了眉目。包括近期被热议的 Manus，都在催化整个市场对 AI 应用形成共识。

知名咨询公司 a16z 月初发布了最新的全球 AI 应用排行，基于 Sensor Tower 的月活跃用户数据。距离上一次排名更新仅过去 6 个月，已经能看到格局大洗牌。有发布 20 天内突破 1000 万用户的 DeepSeek 上榜，也有豆包、百度、美图等国内应用排名变动。

全球 AI 应用排行（2024.8/2025.3）；来源：a16z

在当下这个时间节点，AI 应用的开发门槛降低，结合消费需求去定制产品的自由度变大，过去暂时被搁置的商业化效率，权重也在上升。

a16z 的统计中，40% 高收入应用未进入用户量 Top50。最受消费者喜爱的应用，不一定具备专业用户愿意付费的高阶、或者企业级功能。

当 AI 真正开始有技术普惠的迹象，把材料转化为成品，就是考教各方“锻造”技术的时刻。当下，也确实出现了一些值得参考的案例。

就像即时通讯技术早已成熟，但直到微信诞生以后，才真正“一统江湖”。底层技术的突破与精研没有止境，但现在，该轮到产品经理发力了。

01 高维技术的普惠以“下沉”为前提

“认知”是消费绕不开的前提，作为消费品的应用同理。

产品导向型的营销观念认为只要技术力到位，产品足够优秀，就能靠口碑口耳相传吸引消费者。但现代营销不是“酒香不怕巷子深”的年代，用户日均接受成百上千条广告信息。2022 年的一份报告指出，当代人的连续专注时长，已从 2000 年的 12 秒下降到了如今的 8 秒，和金鱼不相上下。

AI 应用更特殊，因为存在双重认知阻碍。首先是技术理解层面，绝大多数用户对深度学习、AIGC 的运作逻辑没有概念。再加上，AI 所提供的服务存在一定感知滞后，它不像通讯工具（如微信）解决即时沟通这样的显性需求，也非娱乐应用直接提供感官刺激。在没有现实经验迁移的前提下，AI 应用的市场教育难度更大，推广极其依赖热启动。

B 端产品经理如何快速成长？

产品与业务架构主要是将整个业务工作流进行分层，梳理，然后抽象出一个个需求，将业务需求与产品合情合理的映射起来，最终使业务数据在产品中流动，执行，记录，使用。

查看详情 >

因此，背靠抖音和微信，有原生流量池的豆包、元宝，推广路径最为典型。

豆包把“下沉”做得更透彻，它的“与众不同”从第一眼就能分辨。AI 应用命名有三个派别，或是文心一言（文小言），通义千问这类在中文里引经据典，或是 Kimi、DeepSeek 这种别有意涵的英文，还有“豆包”。当一众 AI 应用图标追求设计上的极简、科技感，豆包选择了一个极其“亲民”又 AI 味十足的卡通大头人像。

背靠字节强大的推广渠道，豆包迅速实现了用户规模断层。天风证券报告显示，去年豆包 4 月和 5 月投放金额近 1800 万元，到 6 月上旬骤升至 1.24 亿元，其中 89% 的广告投放比例投入至巨量引擎，字节还限制 Kimi 等竞品在抖音平台的广告投放。

传播学有个“数字鸿沟”的理论，指因媒介素养的不同，造成信息落差两极分化，比如中老年人群，经常被数字鸿沟挡在新技术门外。凭借短视频在各类圈层的触角，豆包可能是“泛下沉”意义上认知度最高的 AI 应用。它的策略是把技术高冷的外衣换成卡通形象，让“AI 带娃”、“春节盲盒”这些话题融入日常议程，专业性还在次，先让用户有交互获得感。

腾讯对 AI 应用侧的反应或许有些后知后觉，但腾讯元宝近期的投放力度打出了后发先至的声势。

元宝去年 5 月上线，早期在一众通用型 AI 应用中存在感并不强，那时候豆包、Kimi 等 App 的投流大战已经开始。但 DeepSeek 爆火后，腾讯一反守势全力进攻，策略也很直白，就是借势宣传，把“搭载满血 DeepSeek”写在名称后缀上。

AppGrowing 数据显示，元宝 30 天内的预估投放金额已超 5 亿元。除了在各个渠道大规模投放广告，腾讯自家生态内的推广几乎被用到极致，包括微信生活服务“九宫格”这一黄金推荐位。

如果说豆包是一开始就下定决心走“强推”路线，元宝势必要弯道超车的架势，除了受到当前市场情绪高涨的影响，也因为 AI 应用的心智之争不会再给落后的玩家太多时间。更何况，获取认知度只是第一步。

02 易用尚且不等于实用

QuestMoblie 年初的报告显示，去年下半年综合类原生 App 在使用粘性上提升明显，然而大部分依然面临卸载率高的难题，AI 的价值尚未完全发挥。也是因为高卸载率，各家才需要不断拉新、继续投流，从而稳固市场份额。

来源：QuestMoblie

进入产品质量评测层面，“易用”大概是第一道门槛，它最直观的评价标准是，要尽可能避免中间商赚差价。

倒卖 AI “知识”获利几乎是 ChatGPT 爆火后经久不衰的新闻，只不过主角随版本更新，最近换成了 DeepSeek。由于技术门槛颇高，相关知识付费需求旺盛，在借助技术迭代的话题催化一些焦虑情绪，大量“信息差”套利的机会因此诞生。

进入应用遍地开花的时代，很多产品为了提供差异化服务，都在易用性上下了功夫。比如提供定制化的场景，调教好的智能体，可以直接套用的模版，避免“Prompt 鸿沟”。核心逻辑是，减少用户完成目标所需要的操作步骤和学习成本。

易用对于消费级应用的意义，可以参考另一个案例。PhotoShop 统治了图形处理工具十几年，其催生的大量设计课程，外包中介也繁荣了十几年。但 Canvas 通过网页入口、模板化操作，大幅降低了使用门槛，让用户能在有限的自由度下完成设计。这就是最典型的，技术底座和产品设计在目的性上的差异。

但更进一步的，AI 生成内容能否在真实应用场景中顺利交付，各方尚处于混战阶段。

最显而易见的问题，怎么在同质化的产出中脱颖而出。现阶段的 C 端 AI 应用可大致分为原生独立 App 和插件（嵌入原有 App）两种，其覆盖场景横跨教育学习、休闲娱乐、图像生成、生活助手、情感陪伴等等大类。

假如我是一个自媒体文案，想给社交平台上每日定期更新的内容设计文案，我可以找到 5 个以上能实现这个目标的 AI 工具，有些甚至提供各类场景模版，再进一步，已经有内容平台可以 AI 一键生成文案，省去“外包”的程序。

“实用”的判定逻辑是，系统能在真实场景中稳定、可靠地输出符合预期的结果。现阶段，用户使用 AI 应用的大部分时间都在做这个判定动作，且显而易见的是，没有断层领先的应用。“能用”的底层支撑是技术厚度，DeepSeek 以此成为当前焦点。

不过《新立场》认为，近期赚足关注度的 Manus，抛开营销争议，恰恰是在产品设计上提出了新的角度。

DeepSeek 展示的思考过程让用户直观看到了 AI 的生成逻辑，这是它初期被提及时广受好评的一个点。Manus 一定程度上放大了这种获得感，其产品的交互逻辑是，只需要一个简单的指令，让 AI 去完成从拆解需求，到调用不同 API 一步一步执行任务的过程。

全程可视化，再加上用户可调适，这一定程度上结合了易用性与实用性，虽然具体产出效果还有打磨空间，但确实带来了一些体验感上的革命。

这提示了我们一种可能，作为一项划时代的技术，其落地应用的设计完全可以更进一步，去做一些“重新定义”的，超脱原有范式的尝试。

03 写在最后

据 QuestMoblie，规模在亿级以上的 App 中有 70% 正在“转型路上”，或是以 AI 插件武装核心业务，或是借 AI 探索场景延伸。比如支付宝的生活服务助手支小宝，微博新推出的互动板块微博智搜，以及近两个月席卷整个互联网服务的“接入 DeepSeek”风潮。

这也是为什么我们说 AI 应用真正进入了“搏杀”阶段。当触手可及的各类工具、娱乐都嵌入了 AI 功能，用户对它的感知、评价等等反馈信息将会指数级增长，留存率的分化将会更明显。

现阶段，绝大部分用户仅习惯使用 1 个综合类原生 App，占比到了 80 %以上，愿意使用 4 个以上 App 多方对比的用户仅占 1%，二八定律明显。

换言之，尽管 AI 大概率将成为互联网“水电煤”般的存在，仍然只有少数成熟的 AI 产品能摘取最甜蜜的果实。

*题图及文中配图来源于网络。

作者：王威 编辑：李凡

来源公众号：新立场Pro（ID：xinlichang66），换个角度看世界。

本文由人人都是产品经理合作媒体 @新立场 授权发布，未经许可，禁止转载

题图来自Pixabay，基于 CC0 协议

该文观点仅代表作者本人，人人都是产品经理平台仅提供信息存储空间服务。