The domestic chatbot market has recently experienced significant turbulence, with Doubao achieving over 100 million daily active users and potentially leveraging its sponsorship of the 2026 CCTV Spring Festival Gala for further expansion. Yuanbao has pursued growth through a 10-billion-yuan red envelope campaign and the social innovation feature Yuanbao Pai. Meanwhile, Alibaba's Tongyi Qianwen has concentrated on developing Agent capabilities to serve user needs through e-commerce and local services. How should we interpret the competitive landscape of domestic chatbots?
We believe the competitive structure of the domestic chatbot market may have already taken initial shape. We reiterate the assessment from our "Chatbot Topic: The Future Has Arrived" report: given Doubao's unique product positioning of "starting with EQ, while being compatible with IQ," along with its comprehensively leading data in user scale, stickiness, and retention rates, other applications will likely find it difficult to significantly narrow the gap with Doubao's scale.
Starting with EQ and being compatible with IQ represents Doubao's distinctive positioning. If we analogize a large model's rational intelligence capabilities, such as programming, to its IQ, and define dimensions like emotional companionship and encouragement as its EQ, then Doubao relatively emphasizes "emotion and EQ" across aspects like human-computer interaction methods, product definition, and model capabilities. This has become its core differentiation and advantage compared to competitors.
User data proves that Doubao's positioning effectively addresses core user needs. According to LatePost, Doubao's DAU surpassed 100 million in December 2025, notable as it is a product for which ByteDance has historically exercised relative restraint in user acquisition spending. QuestMobile data shows that Doubao's DAU/MAU ratio exceeded 30% in December 2025 and continues on an upward trend, significantly leading the approximately 15% level of other competitors. Doubao's 30-day new user retention rate is also markedly higher. This data demonstrates that Doubao's unique "start with EQ, compatible with IQ" positioning resonates with core user demands.
Catching up in EQ capabilities presents considerable difficulty. From a product perspective, current competitors have not significantly emulated Doubao's unique positioning (raising the question of willingness to follow). Furthermore, Doubao's high emotional intelligence is essentially the combined result of optimizations across models, product, engineering, and big data, forming a competitive barrier. Competitors aiming to catch up must do so systematically, unable to achieve a focused, breakthrough-style catch-up similar to Gemini's approach relative to ChatGPT (which relied on advances in model intelligence). The objective difficulty of catching up is high. Based on this, we suggest monitoring the subsequent development of Doubao's overseas version, Dola, which also holds significant potential.
Risks include uncertainty in the development path of large models; unpredictability in product commercialization; regulatory and geopolitical risks; and ambiguity in the evolution from Chatbot to Agent.
Event: Chatbots Engage in Spring Festival "Red Envelope Battle" From December 2025 to the Spring Festival in February 2026, a new round of competition centered on AI assistants fully unfolded. Unlike previous contests focused primarily on large model capabilities, the core of this round lay in securing actual user access points and scenarios. The three major internet companies—ByteDance, Alibaba, and Tencent—leveraged their respective core strengths to choose differentiated competitive paths, concentrating their efforts around the critical period of the Spring Festival.
Phase 1: Organizational and Strategic Positioning (December 2025) In early December 2025, Alibaba Group merged relevant business lines to formally establish the "Tongyi Qianwen C-end Business Group." On December 17, 2025, Tencent upgraded its large model R&D structure, appointing former OpenAI researcher Yao Shunyu as Chief AI Scientist. On December 24, 2025, ByteDance's AI assistant "Doubao" saw its daily active users (DAU) surpass the 100 million mark, becoming the first standalone AI application in China to achieve this milestone. On December 29, 2025, ByteDance officially announced that its Volcano Engine would become the exclusive AI cloud partner for the 2026 China Central Television (CCTV) Spring Festival Gala, with its AI assistant "Doubao" simultaneously launching interactive activities during the gala.
Phase 2: Product Feature Releases and Spring Festival Preparations (January 2026) On January 15, 2026, Alibaba's "Tongyi Qianwen" was fully integrated into ecosystem apps like Taobao, Alipay, Amap, and Fliggy, launching over 400 "AI Task" functions. Users could directly complete operations like booking tickets, shopping, and ordering food via natural language commands, with in-app payment capabilities. On January 26, 2026, Tencent unveiled its "Social + Red Envelope" Spring Festival strategy, with "Yuanbao" previewing two key actions: 1) launching the innovative social feature "Yuanbao Pai," supporting group chats and synchronized entertainment interactions between users, friends, and AI; 2) announcing the "Use Yuanbao, Share 1 Billion" cash red envelope campaign starting February 1. Pony Ma internally expressed hope to replicate the success path of WeChat red envelopes achieved through Spring Festival social fission. On January 29, 2026, at ByteDance's all-hands meeting, CEO Liang Rubo set the company's annual keyword as "Scaling New Heights," explicitly stating that the "short-term peak is excelling in the Doubao/Dola assistant applications." This directive signifies a high-intensity concentration of company resources towards the AI assistant business, responding to market competition pressure.
Phase 3: Spring Festival Showdown and Direct Confrontation (February 2026) On February 1, 2026, Tencent's "Yuanbao" 10-billion-yuan red envelope campaign launched as scheduled, rapidly disseminating through social networks like WeChat and QQ. On February 2, 2026, Alibaba announced that the Tongyi Qianwen App would initiate a "Spring Festival Treat Plan" with investments up to 3 billion yuan, scheduled for official launch on February 6. This plan will involve core businesses within the Alibaba ecosystem like Taobao Flash Sales, Fliggy, Damai, Hema, Tmall Supermarket, and Alipay. On February 3, Alibaba further announced exclusive title sponsorship of the Spring Festival galas of four provincial satellite TV stations: Dragon TV, Zhejiang Satellite TV, Jiangsu Satellite TV, and Henan Satellite TV.
How should we view the competitive landscape of the domestic chatbot market? We reiterate the judgment formed in our "Chatbot Topic: The Future Has Arrived" report: Chatbots have become the "Killer App" of the AI application era and possess the potential to evolve into super-entrances for comprehensive Agents, holding significant strategic importance for internet giants. This is the core reason recent related companies have utilized the approaching Spring Festival, employing methods like red envelopes to attract users. Regarding the competitive landscape, based on Doubao's unique "start with EQ, compatible with IQ" product positioning and its comprehensively leading user data in scale, stickiness, and retention, we believe the domestic chatbot competitive structure may have already initially solidified. Short-term, elements like red envelopes and social features may aid user acquisition, but long-term competitive dynamics will ultimately be determined by the product's core competitiveness, which dictates user retention.
What does "Start with EQ, Compatible with IQ" mean, and why is Doubao's product positioning considered unique? We analogize with human concepts of IQ and EQ: if a large model's rational intelligence capabilities like logic, programming, and mathematics are analogous to its IQ, then in an IQ context, a Chatbot resembles more of an efficiency tool. Furthermore, we define dimensions like emotional companionship, friendliness, and encouragement as a large model's EQ. In an EQ context, the Chatbot resembles more of a companion or a trustworthy friend. We believe that unlike overseas products like ChatGPT and Gemini, which focus more on IQ, Doubao's unique positioning抢先 focuses on EQ. Evidence suggests that while the increasingly advanced intelligent capabilities of large models are a necessary step towards AGI, top-tier abilities like programming and mathematics may not be essential needs for most ordinary C-end users. Conversely, humans are inherently emotional beings; a high-EQ companion has been validated as an application scenario better aligned with the needs of the majority of ordinary users.
Why is Doubao's positioning considered unique? We demonstrate Doubao's unique "start with EQ, compatible with IQ" positioning from product and user data dimensions. Firstly, on the product side, unlike ChatGPT's core interface resembling a concise, search-like efficiency tool, Doubao, upon its launch in 2023, drew significant inspiration from Character.ai, a tool inherently positioned for AI chatting. According to LatePost: Doubao's lead, Zhu Jun, holds rather differentiated imaginations for AI—while ChatGPT seemed to define the AI assistant product form, he insisted that AI should be more "anthropomorphic" and "human-like." Additionally, Doubao is the only mainstream Chatbot product with an anthropomorphic image; its default voice and avatar is "Peach."
Text itself is relatively rational; emotional transmission relies more on multimodal capabilities. Doubao fully leverages ByteDance's advantages in multimodal large model capabilities and its own product engineering prowess to achieve high emotional intelligence. From a model capability perspective, ByteDance's Doubao series models do not hold a significant advantage over competitors like DeepSeek and Tongyi Qianwen in the IQ dimension. However, in the multimodal dimension, particularly in end-to-end voice models, Doubao早在 launched its Voice 1.5 version (January 2025), where its chat anthropomorphism level surpassed GPT-4o. Conversational interaction represents a new human-computer interaction paradigm in the large model era, serving as a significant supplement or even upgrade compared to traditional multitouch. On the other hand, voice conversation has also become an important usage scenario for emotional companionship and other EQ dimensions.
Furthermore, Doubao conducts deep product optimizations based on different users' EQ needs. According to LatePost: "Doubao launched a real-time voice call function, with 'emotion' being a key focus—they traveled across the country to collect dialects, refining accent granularity down to the district/county level within cities; they also conducted a series of stylized dialogue training for the model, endowing it with distinct personality." We compared the conversational chat capabilities of Doubao, Tongyi Qianwen, and Yuanbao, finding Doubao significantly领先 in dimensions like the number of usable voices and dialects. Tongyi Qianwen's conversation feature is placed under a deeper entry point, indicating it is not a core function. Overall, we believe Doubao's unique "start with EQ, compatible with IQ" positioning is not fully reflected in other competitors like Yuanbao, Tongyi Qianwen, or even overseas products like ChatGPT and Gemini; Doubao currently maintains significant differentiation.
Starting with EQ does not equate solely to emotional companionship chat, nor does it mean abandoning IQ. High emotional intelligence is Doubao's unique positioning and advantage, but it does not represent its only attribute (it has not舍弃 IQ). As an efficiency tool and in the IQ dimension, it remains not inferior to main competitors. In terms of comprehensive large model intelligence comparison, Doubao still ranks within the domestic first tier. Comparing core efficiency tool functionalities, Doubao is also not落后, even leading in some areas. On the other hand, even within the EQ dimension, we do not believe emotional companionship chat is the only scenario; high EQ can be integrated into every user query and interaction, understood as a way of interacting with users that helps build trust and stickiness.
Doubao's unique positioning stems primarily from ByteDance's embrace of innovation and its methodology of decision-making based on user data. Facing an unknown AI product roadmap, ByteDance & Doubao are among the major players more willing to allocate resources, tolerate trial and error, and explore前瞻. Initially, Doubao experimented with various possibilities. The essence lies in the fact that in the AI era, the correct path is unknown; a reasonable, objective approach is to persist in experimentation and explore possibilities. Consequently, Doubao's marketing expenses are relatively low among ByteDance's products because it does not presuppose a路线, nor is it about比拼执行力 to promote a pre-defined mature product. On the other hand, generative AI is fundamentally an industrial transformation driven by technological leaps. Common product thinking often starts from the technology itself to find application scenarios,容易犯 the error of "looking for a nail when you have a hammer." In contrast, ByteDance's core product methodology is AB testing attribution based on user data. Doubao's unique "start with EQ, compatible with IQ" positioning did not originate from a "top-down" preset but from genuine feedback in its user data. This再次证明了 that user demand determines the core product路线, not technology.
User data proves the EQ路线 effectively addresses core user needs in the C-end market. Doubao not only leads significantly in user scale but also exhibits断档式领先 in user stickiness and retention duration. According to LatePost, Doubao's DAU surpassed 100 million in December 2025, and it is a product for which ByteDance has historically been relatively克制 in user acquisition spending. Furthermore, QuestMobile data indicates Doubao's DAU/MAU ratio exceeded 30% in December 2025 and continues to rise, significantly领先 the ~15% level of other competitors. Doubao's 30-day new user retention rate is also markedly higher. Analyzing usage duration distribution, Doubao has a significantly lower proportion of shallow users (0-3 seconds) compared to Yuanbao and Tongyi Qianwen, while its proportion of deep users (over 10 minutes) is higher. If user scale can be attributed to ByteDance's channel traffic advantage, metrics like user stickiness and retention argue for Doubao's unique differentiation—it appears to satisfy user needs more deeply and effectively. Finally, comparing QuestMobile user usage timing distribution between DeepSeek and Doubao reveals that on weekdays, the former's usage peaks during morning and afternoon work hours, whereas Doubao's peak occurs during evening leisure hours. This validates that the former is more偏向 an efficiency tool, while the latter aligns with the "start with EQ, compatible with IQ" positioning.
The overseas chatbot market also shows similar user demand trends, with ChatGPT evolving from an efficiency tool towards a lifestyle. Sensor Tower's "2025 AI App Market Insight Report" indicates ChatGPT's development aligns with this described trend. From Q2 2024 to Q2 2025, the proportion of "productivity" related conversation topics (like work and education) in ChatGPT decreased from nearly 50% to 37%, while the proportion of lifestyle scenarios like health, finance, travel, and entertainment increased from 22% to 35%.
How difficult is it for competitors to catch up via the EQ路线? Based on the analysis in Chapter 1, we believe this results from the interplay of multiple subjective and objective factors: on one hand, it involves strategic recognition and路线认同 (willingness to follow); on the other, catching up is also a "systematic project" difficult to achieve through single-point breakthroughs (ability to follow). Subjectively, Doubao's competitors have not significantly emulated its unique positioning currently, possibly due to issues of strategic recognition and路线认同, or path dependency on their own ecosystem advantages (Tongyi Qianwen emphasizes "getting things done," Yuanbao focuses on "social"), or traditional definitions of "intelligence," leading them to potentially not fully prioritize "emotionalization" subjectively and thus not跟进全力. Objectively, even if competitors decide to catch up, "high emotional intelligence" is a "systematic project" difficult to breach via single-point breakthroughs. Doubao's high EQ is essentially the combined result of optimizations across models, product, engineering, and big data, forming a competitive barrier. Competitors must catch up systematically, unable to achieve a focused, breakthrough-style catch-up类似 Gemini's approach relative to ChatGPT (achieved through model intelligence advancement). Furthermore, if Doubao enters a data growth flywheel based on its user scale advantage (more user data leads to better user experience), similar to internet industry patterns, the gap between the leader and other participants could even widen.
Additionally, based on current AI model training architectures, reinforcement learning is a key algorithmic mechanism for enhancing model capabilities. The objective function for the high-EQ路线 is user feedback, which provides a relatively clear target. Even within existing architectures, Doubao's EQ has significant room for improvement. Conversely, for some前瞻, systematic thinking capabilities, current large model architectures lack a clear emotional objective function; perhaps breakthroughs in underlying algorithmic mechanisms are needed to achieve true AGI.
Tracing the source, Doubao's lead本质上 stems from ByteDance being among the major players most resolute in its commitment to AI assistants, with greater tolerance for error and more aggressive exploration. In an era where the absolutely correct AI path is unknown to anyone, its early "hodgepodge" style exploration was恰恰 a reasonable strategy. Therefore, its marketing expenses are relatively克制, as the goal was not to promote a预设 mature product. In contrast, some competitors may exhibit strategic矛盾: initially, when the direction was unknown, investment in trial and error was insufficient, and exploration lacked resolve; later, upon observing a certain path (like red envelope marketing, ecosystem integration), they tend to认定 it as the "correct路线" and invest in饱和式 marketing. This pattern of "conservative exploration前期, aggressive marketing后期" may reflect unclear thinking about what unique value the product itself should possess, or an urgency to establish a position within the window period, attempting to substitute fundamental product innovation and路线 exploration with execution and resource投入.
How should we view the domestic chatbot Spring Festival "red envelope shower"? Short-term disruptions are unlikely to撼动 long-term chatbot格局 judgments. Simply put, we believe these elements do not strike at the core of product competitiveness and are more short-term扰动 factors. If competitors cannot significantly narrow the user experience gap in terms of core product competitiveness, even if red envelopes aid user acquisition, user retention remains questionable. In fact, the extent of these short-term扰动 might even be limited. For example, according to DataEye data, from November to December 2025, apps like Tongyi Qianwen and Yuanbao had significantly higher exposure in the user acquisition market compared to Doubao, yet Doubao held the #1 position on the iOS download chart for most of that period. Yuanbao recently surpassed Doubao in rankings leveraging its Spring Festival red envelopes. On January 29, at ByteDance's first all-hands meeting of 2026, CEO Liang Rubo set the keyword for the new year as "Scaling New Heights." In the short term, the peak is "the Doubao/Dola assistant applications," highlighting the current importance of Doubao within ByteDance's overall AI strategy. Subsequent actions by Doubao regarding Spring Festival marketing are also值得期待. On the other hand, the Dola mentioned by Liang Rubo is the overseas version of Doubao. According to LatePost, its DAU has already surpassed 10 million. We believe it also has the potential to carve out a space in overseas markets by借鉴 Doubao's similar unique "start with EQ, compatible with IQ" positioning.
What is Dola? Dola is a free AI assistant application developed by ByteDance specifically for overseas markets, formerly known as Cici launched in August 2023, and rebranded in late 2025. Dola's interface resembles Doubao, also featuring a rich set of official agents, support for creating and sharing personalized agents, and a considerable selection of anthropomorphic voices. Its image generation function supports multiple templates. User feedback from channels like the Apple App Store indicates that image generation capability is a highly praised feature of Dola, with users generally considering its AI drawing quality high, rich in detail, accurately preserving facial features from the original image, comparable to or even surpassing some paid tools, and easy to operate. Furthermore, user evaluations suggest that compared to some competitors requiring multi-step verification, Dola is more user-friendly, usable directly without complex registration or downloads. However, compared to Doubao domestically, which has begun attempting to divert traffic to Douyin Mall and even explore system-level "phone assistant" features, Dola's deep integration with e-commerce, lifestyle services, and other ecosystems overseas has not yet been scaled, and its functionalities remain relatively focused on basic assistant services.
By late 2025, its daily active users (DAU) had surpassed 10 million, establishing it as a second-tier AI application with a stable user base overseas (comparable in scale to Grok and Character.ai). According to Sensor Tower, as of December 2025, Dola's total MAU reached 55.93 million.
Replicating ByteDance's "emerging markets first" globalization strategy. It avoids direct confrontation with giants like OpenAI and Google in North America, instead focusing on regions like Indonesia and Mexico where AI assistant penetration is still low and mobile internet growth is rapid, capturing the market through a free, lightweight, and highly practical product strategy: Since October 2025, Dola一度 ranked within the top 20 free app downloads on the Google Play Store in Indonesia, Malaysia, Mexico, and the UK consecutively. Additionally, Dola actively expanded into Latin American markets, notably Mexico, where it topped the Google Play Store's daily download chart for a consecutive week. To date, its MAU in Mexico has reached 8.29 million. Building on steady growth in Southeast Asia and Latin America, Dola is actively penetrating developed markets (e.g., Dola's MAU in the UK has surpassed 1 million), indicating that its overall growth still possesses vast potential.
Risk Reminders Uncertainty in the development path of large models: Although leading players like OpenAI, Google, and ByteDance continue to make breakthroughs in foundational model capabilities, reasoning performance, and multimodal directions, overall, large models and Agents remain in a phase of rapid iteration. Their stability, interpretability, cost controllability, and generalization ability in real-world complex scenarios are not yet fully mature. On one hand, models may still experience hallucinations, execution deviations, or failures during multi-step decision-making and real-world execution (e.g., cross-system operations, transaction closures). Furthermore, the underlying technological paths and architectural choices for generative AI are not fully finalized, leaving room for iteration or shifts. Additionally, this industry is technology and capital-intensive, requiring continuous large-scale R&D and computing power investments with long return cycles. Simultaneously, due to the relatively high degree of open-source core technologies and intense competition from numerous tech companies, industry entry barriers are somewhat limited, sustaining fierce market competition. The combined uncertainties in technology, investment, and competitive environment may impact the long-term stable development of specific companies or technological paths.
Uncertainty in product commercialization: Currently, companies are exploring diverse commercialization paths including subscription & API fees, cloud & platform enablement, advertising & e-commerce conversion, and enterprise-level solutions. However, profit models overall remain in the validation stage, presenting significant commercialization uncertainty. Furthermore, industry price wars could compress profit margins, and the user stickiness and payment willingness of existing Chatbots and Agents may face fluctuations.
Regulatory and geopolitical risks: AI-native applications directly interact with end-users and core business operations, involving multiple regulatory issues like data privacy, content security, and algorithm transparency. Scrutiny over large model training data sources, output content liability, and system-level入口 might intensify, potentially imposing constraints on model training, product functionality, and cross-border data flows. Simultaneously, high-end computing chips and cloud infrastructure are also subject to geopolitical influences and export controls, potentially leading to significant increases in computing costs and slower model iteration.
Uncertainty in the evolution from Chatbot to Agent. The evolution from Chatbot to Agent is not a smooth functional upgrade but a leap involving technology, product, commerce, and ecosystem standards. While industry consensus points towards Agents capable of autonomously understanding, planning, and executing complex tasks, their technical maturity (e.g., long-term memory, reliable reasoning), user acceptance based on trust, and clear business models are not yet formed. Additionally, whether users are willing to delegate tasks involving privacy, payments, or critical decisions to Agents, and accept potential erroneous outcomes and responsibility ambiguities, presents significant hurdles in habit cultivation and trust building. For current leaders, their existing success is built on the logic of the Chatbot era; whether they can continue to lead in the Agent era involves considerable uncertainty.
Comments