From Chatbot to Local Life-Service Gateway

ByteDance’s Doubao is no longer just an ordinary AI chatbot. After integrating with Douyin e-commerce and supporting in-app shopping orders, Doubao has further connected with Douyin Life Services. Users can ask questions in natural language, receive store and package recommendations, complete group-buying purchases, pay inside the app, view orders, and retrieve verification codes for offline redemption. The currently supported categories include food, movie tickets, and homestays.

The significance of this move lies in how it shows Chinese AI assistants evolving from “answer engines” into “transaction interfaces.” In the first stage of generative AI, users opened AI assistants to ask questions, write, translate, summarize, or brainstorm. In the next stage, assistants need to understand user intent and help complete real-world tasks. Local life services are a natural testing ground: people often ask where to eat nearby, which movie to watch, where to stay for a short trip, or which nearby deal is worth buying.

For ByteDance, this is not only a product upgrade, but also a monetization experiment.

The Monetization Challenge for AI Assistants

AI assistants have a very different cost structure from traditional internet products. Social feeds, search result pages, or e-commerce listings can scale under relatively predictable infrastructure costs. Large models, however, consume computing resources every time they respond. Every prompt, every token, and every multi-turn conversation involves GPU, electricity, bandwidth, storage, and engineering costs.

The problem is that users’ willingness to pay remains limited. According to data cited from Tencent Research Institute, 95.6% of respondents had used large-model AI products in the past three months, and 82.3% were high-frequency users, but only 9.8% had paid for such products.

This explains why mainstream AI products cannot rely solely on subscription revenue. Even if users use AI frequently, they may still expect AI services to remain free or low-cost. Advertising is another monetization option, but it carries higher risks in AI assistant scenarios. Users trust AI assistants because they expect answers to be useful, objective, and context-aware. If recommendations look too much like paid advertisements, that trust can quickly weaken.

Therefore, “service distribution + transaction commission” becomes more attractive. Instead of directly inserting ads, an AI assistant can recommend suitable services when users have already expressed purchase intent and guide them toward completing a transaction. If the recommendation is accurate and the transaction flow is smooth, the platform has a chance to generate revenue without significantly damaging the user experience.

The Strategic Value of Local Life Services

Local life services are different from ordinary e-commerce. In standard online shopping, users mainly compare products, prices, reviews, delivery, and return policies. Local life decisions are much more context-dependent. A restaurant recommendation may depend on distance, cuisine, budget, taste, parking, queue length, time of day, number of people, and even whether it is for a date, family dinner, or gathering with colleagues.

This is exactly where AI can create value. Traditional search interfaces require users to filter results manually, while AI assistants can turn vague needs into structured decisions. For example, “Find a quiet restaurant nearby that is suitable for three people, not too expensive, and still open tonight” is not just a keyword query. It contains location, time, price sensitivity, atmosphere preference, group size, and consumption intent.

Doubao’s advantage lies in scale. QuestMobile data shows that as of March 2026, Doubao had reached 345 million monthly active users. Unlike passive content feeds, Doubao receives active user requests. When users ask where to eat over the weekend, which movie ticket to buy, or where to stay for a short trip, their purchase intent is usually stronger than when they casually scroll through short videos.

Therefore, ByteDance hopes to turn Doubao into an “AI life-service platform.” The goal is not only to recommend content, but also to capture service distribution and transaction conversion opportunities at the moment users express real needs.

The “De-Search” Logic: Bypassing Meituan’s Barrier

Douyin has already reshaped content commerce through the “goods find people” model. Short videos and livestreams can stimulate demand before users actively search. However, this model also has limitations. In e-commerce, impulse purchases may bring return pressure; in local life services, users may buy group deals but never redeem them. The article also points out that traffic cannot be consumed endlessly, and Douyin is trying to strengthen search and shelf-based scenarios to obtain more certain transactions.

However, in search-based local life services, Meituan has accumulated advantages over many years. Users have already associated Meituan with restaurants, reviews, group buying, food delivery, and local service comparison. Competing directly through a search entrance is not easy.

AI provides another path: de-search. Doubao does not necessarily require users to actively search, compare, filter, and decide. Instead, it can guide the decision-making process through conversation. The assistant can become the interface between user intent and service inventory. If this path works, ByteDance will not need to compete with Meituan only through a traditional search box. It may have the opportunity to build a new habit: ask AI first, then buy.

For developers, multi-model calls, cost control, and service stability will become important engineering issues for local life AI applications. In real projects, development teams can also choose to integrate a large-model API aggregation platform such as TreeRouter according to business needs, using it as a supplementary solution in the model service call chain, so that they can more flexibly adapt to different model capabilities across various scenarios.

Developer Perspective: A Business Funnel Example

The core question is whether a massive AI user base can truly be converted into a reliable transaction engine. The following sample code uses key data from the article to illustrate scale sensitivity. This code is for demonstration only and does not represent Doubao’s actual paid conversion rate, but it helps developers understand that even a relatively low conversion rate can correspond to a considerable user scale when the monthly active user base is extremely large.

doubao_mau = 345_000_000

survey = {
    "used_ai_models_last_3_months": 0.956,
    "high_frequency_users": 0.823,
    "paid_users": 0.098,
}

def estimate_users(base_users: int, ratio: float) -> int:
    return int(base_users * ratio)

estimated_paid_scale = estimate_users(doubao_mau, survey["paid_users"])
estimated_high_frequency_scale = estimate_users(doubao_mau, survey["high_frequency_users"])

print(f"Doubao MAU: {doubao_mau:,}")
print(f"Hypothetical paid-user scale at 9.8%: {estimated_paid_scale:,}")
print(f"Hypothetical high-frequency-user scale at 82.3%: {estimated_high_frequency_scale:,}")

The possible output is as follows:

Doubao MAU: 345,000,000
Hypothetical paid-user scale at 9.8%: 33,810,000
Hypothetical high-frequency-user scale at 82.3%: 283,935,000

The point of this code is not simply to argue for raising subscription prices. Instead, it shows that the key to AI monetization lies in finding high-frequency scenarios and forming a closed loop among user intent, trust, and transaction infrastructure.

Core Challenge: Recommendation Accuracy and User Trust

Doubao’s challenge is not whether users can technically buy group deals inside an AI assistant. The real question is whether users are willing to trust AI enough to make local consumption decisions.

Local services are highly complex. Two restaurants with the same rating may deliver very different real-world experiences. A hotel may look good in photos but provide poor service. Whether a movie ticket package is truly worthwhile also depends on whether the showtime, seat, location, and discount match the user’s needs. For AI recommendations to truly work, they require structured data, real-time availability, location understanding, user preference memory, and sufficiently strong ranking logic.

The friction points mentioned in the article are also worth noting:

  1. After Doubao connected with Douyin Life Services, it still does not fully enjoy the subsidies available in Douyin’s main app. The same package may have different prices in Doubao and Douyin, which could push users back into price comparison.
  2. Compared with Alibaba’s Qianwen, Doubao currently focuses more on recommendation and selection, while users still need to complete orders themselves.
  3. The payment infrastructure still has room for improvement. Douyin Pay still lags behind Alipay in terms of user coverage and service depth.

These details directly affect the user experience. If an AI assistant can only “make recommendations,” the experience remains experimental. Only when recommendations are accurate, prices are consistent, payments are smooth, and after-sales service is reliable will users truly regard it as an entrance to local life services.

Conclusion: AI Assistants Need More Than Traffic

Doubao’s integration with life services sends an important signal to the AI industry: the next stage of competition will not only be about model intelligence, context length, or generation quality, but also about whether AI assistants can become trusted commercial interfaces.

ByteDance has a strong starting point: Doubao’s 345 million monthly active users, Douyin’s content and service ecosystem, and its ability to shape user habits. But traffic alone is not enough. To convert AI conversations into real transactions, Doubao must solve problems around recommendation accuracy, price consistency, payment smoothness, service reliability, and user trust.

If this path works, AI assistants may become a new entrance to local life services. If it does not, users may still use AI for suggestions, but ultimately return to Meituan, Douyin’s main app, or other mature platforms to complete purchases. The real test is whether Doubao can evolve from “help me choose” to “help me finish it.”