American scholars: Chinese enterprises are taking a smarter path in the field of AI

While American AI giants are burning money on Super Bowl ads, Chinese AI is using subsidies to cultivate user habits. After comparing the two approaches, American scholars concluded that Chinese companies are taking a smarter route – instead of competing on model parameters, they are embedding AI into things users do every day. Capability wins trials, and habit wins it all.

Earlier this year, tech giants in China and the United States launched two very different marketing campaigns to promote their new AI products.

During the Chinese Spring Festival, Alibaba distributed over $400 million in subsidies to its customers. Consumers could get subsidies for every meal, movie ticket, flight booking, and fresh food delivery completed through Alibaba's AI assistant Tongyi Qianwen – as long as the entire transaction was handled by Tongyi Qianwen.

Two weeks later, at the 60th Super Bowl on the other side of the Pacific, OpenAI, Google, Meta, and Amazon each spent up to $10 million to buy 30 - second ad slots, trying to convince American consumers that their AI was the smartest, most capable, and most transformative.

The differences between these two campaigns reflect that companies in the two countries are adopting very different strategies. More and more evidence shows that at least for now, Chinese companies' choices are wiser.

US Strategy: Competing on Capability

In the United States, the mainstream AI strategy is to compete on capability: create a competitive advantage through better training data, larger models, better benchmark performance, and more advanced features. The assumption is that capability leads to user adoption, and user adoption leads to market dominance. The AI interface itself becomes a destination, a place where consumers actively go for research or transactions.

This logic is deeply ingrained in technology strategies. In previous platform battles – search engines, social networks, mobile operating systems – the best products usually won and maintained the lead for a long time.

Competing on capability has indeed driven the growth of the AI industry. Just ChatGPT alone saw its weekly active users double in 2025, reaching 800 million, and the entire market has been expanding rapidly. Overall, the number of American AI users is growing at a rate of about 30% per year. This is a high - growth industry.

But behind the overall growth, an unsettling pattern has emerged: no American AI company can maintain the lead for a long time. When Google released Gemini 3 in November 2025, OpenAI's CEO Sam Altman issued an internal "red alert" memo and advanced the company's next release by several weeks. And just three years ago, it was Google that issued its own "red alert" in response to ChatGPT.

This pattern is clear: in the AI race, today's market leader is tomorrow's chaser. Every benchmark advantage is temporary. Every new version release triggers panic among all competitors. Companies invest billions of dollars in marginal improvements, only to find that competitors catch up within weeks. The corporate market share tells the same story: OpenAI's share dropped from about 50% in 2023 to 27% in 2025, while Anthropic rose from 12% to 40%, and Google climbed to 21%.

More importantly, the wow factor of each new improvement in AI capability is decreasing. When all mainstream AI assistants can competently answer most questions, write decent emails, summarize documents, and generate usable code, the marginal improvement brought by a "better" model becomes imperceptible to ordinary users. The difference between 90% and 93% accuracy is significant to researchers running benchmark tests, but much less so to a consumer deciding whether to open ChatGPT or just type a question into Google.

This instability and the narrowing capability gap seem to be making consumers lose interest in AI. Our analysis of about 119,000 social media posts during the Super Bowl week (the period with the most intensive AI ad placements in the West) found that the negative sentiment towards AI ads was more than 2.5 times the positive sentiment. When filtering to posts specifically discussing AI ads, the negative sentiment reached 95%, with recurring themes of aesthetic fatigue and accusations that AI ads are normalizing surveillance. Notably, the only AI - related ad that made it into the top ten of audience favorites was Ring's "Search Party" dog - finding feature – this ad didn't mention AI at all and focused entirely on a specific and emotionally resonant result.

In this context, competing on capability no longer seems like a good strategy. So, is there an alternative?

Digging the Habit Moat

Another option is to dig a "habit moat": deeply embed AI into users' daily lives so that switching to an alternative would be laborious, even if the alternative might be slightly better.

Most moats in strategic defense come from scale, network effects, or data lock - in. The habit moat is different: it exists in customers' behavior, not in the company's infrastructure. Behavioral scientists believe that a habit consists of three parts: the signal that triggers the behavior, the behavior usually triggered by the signal, and the reward (the positive result that reinforces the repetition of the behavior). Users don't consciously choose the behavior – the cue ("I want to watch a movie tonight") directly triggers a specific behavior (opening ChatGPT and asking about movies in theaters).

Breaking this sequence requires users to consciously overcome an automatic reaction: stop, notice the cue, and then make a different choice. This is a psychological switching cost, which can be unreasonably high. This is usually why people continue to use banks they don't like, objectively worse browsers, and search engines they no longer prefer – even if the actual operational cost of switching is low.

The formation of online habits requires three conditions: consistent scenarios (the same application, the same time, the same intention), high - frequency trigger cues (the more times users are exposed to them each day, the faster the behavior routine becomes automated), and predictable rewards (users must get what they want quickly every time). Food delivery, ride - hailing, payment, and local services are all online interaction scenarios that meet these conditions, and they are exactly the areas where Chinese AI is being deployed.

Making Users Addicted

Tongyi Qianwen launched its promotion campaign in January 2026, introducing what industry observers call "the world's first comprehensive AI agent super - app", integrating more than 400 capabilities from Alibaba's entire consumer ecosystem: Taobao for shopping, Alipay for payment, Ele.me for food delivery, Fliggy for travel, and Gaode for navigation. The app received 10 million downloads in its first week of public testing. Soon after, the Spring Festival promotion boosted its popularity and was a great success, with about 140 million users completing their first AI - driven shopping experience through Tongyi Qianwen's agent function. By mid - May, this number had reached 300 million.

Tongyi Qianwen represents a concept very different from the American approach. Instead of positioning AI as a high - quality destination that smart users would consciously choose to visit, it views AI as the path users take when doing things they already intend to do. Mark Greeven and his colleagues wrote in Harvard Business Review that China's advantage in the field of agent commerce lies in its infrastructure – the coordinated operation between payment, logistics, and super - apps. They are right about the infrastructure. The deeper question – and the one this article aims to answer – is what possibilities these infrastructures create in users' minds.

Take the user experience as an example. In the past, booking a movie ticket required opening a ticketing app, choosing a theater, selecting seats, comparing prices, and completing the payment, which might take seven or eight steps across multiple pages. With Tongyi Qianwen, the same task is simplified to one sentence: "Book two movie tickets for tonight, middle seats." The rest is handled by AI.

When Tongyi Qianwen was launched, the most eye - catching demonstration involved restaurant reservations: the AI filtered restaurants based on the user's location and preferences, then called the restaurant. Its AI voice was so realistic that the audience wondered if they were hearing a real person or a machine. The AI didn't just recommend restaurants or provide phone numbers. It "picked up" the phone, made the call, negotiated the reservation, and then reported the result.

The strategic insight behind this is deeper than just convenience. Tongyi Qianwen doesn't require users to come to it for specific AI - related tasks. Instead, it intercepts users while they are performing existing tasks and provides a faster way to complete them. The psychological framework shifts from "I should try using AI" to "That was so convenient. I'll do it this way next time." This is the beginning of a habit.

Alibaba's approach is not unprecedented. It is an application of the strategy Tencent formulated a decade ago in the AI era. In 2014, WeChat launched digital red envelopes during the Spring Festival, allowing users to send small cash red envelopes through the app. On the surface, it was just a fun cultural adaptation. In reality, it was a behavioral intervention that trained hundreds of millions of users to link their bank accounts to WeChat and normalized mobile payment. Within three years, WeChat Pay took 40% of the Chinese mobile payment market from Alipay. Tencent spent almost no money on subsidies. It just took advantage of an existing cultural habit and inserted itself into it, becoming the path of least resistance.

Some readers might object that the habit moat is a phenomenon unique to China, and Alibaba's success reflects the unique structure of its existing ecosystem. There is some truth to this. But this objection points to an opportunity, not an impossibility. Since Western consumers' habits are scattered among Amazon, Google, Apple, and many vertical players, the first company to achieve cross - domain behavioral integration in the United States could capture a disproportionate amount of value. The rewards for solving this problem are greater, not smaller, than in China.

So, what is the solution?

Action Guide for the Habit Moat

The structural obstacles faced by Western companies are real, but they are not unchangeable. Here are four actions, arranged from the easiest to implement immediately to the most structurally ambitious. Each applies to both AI companies competing for consumers' default choices and existing enterprises in industries such as banking, retail, hospitality, and healthcare. Whether or not they are involved in the redesign, their customer relationships will face the risk of being intercepted by platforms.

1. Look for Habit Cues, Not Function Gaps

Most product teams' instinct is to ask: What can our AI do that our competitors' AI can't?

The question for the habit moat is: In our customers' daily lives, what existing behaviors can our AI intercept and complete in a simpler way than the existing path?

Starbucks provides an inspiring example. Its Deep Brew AI platform doesn't try to be the smartest AI on the market. It tracks what each customer orders, when, and where, and then uses these patterns to pre - select their regular drinks, recommend pairings, and send notifications during the morning commute. The app places an order before the customer consciously decides to do so.

By early 2026, Starbucks announced plans for an AI ordering assistant that allows users to describe their mood ("Get a boost for a busy morning") instead of browsing the menu. Starbucks intercepted the highest - frequency cue in its customers' lives (the morning routine) and made AI the path of least resistance through this routine.

Every company has similar cues. For a regional bank, it's the moment when customers check their balances – the most frequent banking behavior for most adults. If the bank makes its AI appear automatically at that moment and provides one - click payment, transfer, and budget query functions, it intercepts a habit cue, and it will be difficult for competitors to replace it. For a hotel chain, it's the moment when business travelers arrive: if the hotel makes its AI the default assistant for ground transportation, restaurant reservations, and meeting logistics, it transforms a single transaction (a room) into a continuous relationship (a journey).

2. Subsidize Behaviors, Not Subscription Services

The default approach in the West is to subsidize user access to services: free trials, first - month discounts, and new users getting premium features. This is the approach OpenAI took when it launched its instant checkout feature in September 2025 – the tool allows users to directly purchase products from partner retail enterprises such as Shopify and Walmart within ChatGPT. Registered users get one month of free access to ChatGPT, after which they need to pay a subscription fee. The goal is to make ChatGPT a platform where people can not only discover products but also buy them.

But six months later, according to Forrester data, only about 30 Shopify merchants integrated with the service. In March 2026, OpenAI confirmed that it was abandoning the instant checkout feature because people weren't using it to shop.

As we've seen before, Tongyi Qianwen took a different approach in its marketing. It subsidizes actual purchase behaviors, thus reducing the cost for users to do something they already intended to do. They spend money to let users experience how convenient it is to use the tool and bet that this convenience will be addictive.

Chinese companies' strategies have re - structured the economics of user acquisition. Instead of giving away a month of ChatGPT Premium, OpenAI could have subsidized the first five DoorDash food delivery orders completed through its agent. Instead of bundling Gemini Advanced with Google One, Google could have waived the booking fees for the first three travel bookings completed within Gemini. The cost per acquisition might be similar. But a user who has booked three dinners through an AI agent is likely to have changed their default behavior.

3. Build Environmental Utility, Not Destination - Style Experiences

As mentioned earlier, the trap for Western companies is to view AI as a destination. The strategy of the habit moat is to embed AI as environmental infrastructure into users' existing workflows. The principle is simple but counter - intuitive: any user - facing AI experience that needs to be actively invoked by users is likely to be ignored. The AI experiences that will survive the next three years will be those that don't need to be invoked at all – because they are where the workflow currently takes place.

The comparison between GitHub Copilot and Microsoft 365 Copilot precisely illustrates this point. GitHub Copilot is effective because it occupies the exact position where developers are typing. Microsoft 365 Copilot, although in the same set of applications used by 450 million people, has a paid penetration rate of only 3.3%, mainly because it still exists as a feature that needs to be invoked, rather than a default workflow path. The architectural implication is that to make AI the default path, users should need to actively opt out, rather than actively opt in.

4. Compete on Repeat Transactions, Not First - Time Transactions

Most companies optimize for first - time transaction metrics: registration volume, free trial conversion rate, app installation volume, and first - day popularity. These describe whether a customer has tried your product once. The habit moat is built on what happens after that, i.e., when the same cue triggers, whether the customer will choose your product again without prompting. Capability wins trials, and habit wins the default choice.

The failure of OpenAI's instant checkout feature well illustrates this difference. The feature led to browsing behavior but no purchase behavior, and crucially, no repeat purchase behavior. Users discovered products through ChatGPT but completed transactions elsewhere – at retailers they were already used to. OpenAI optimized for the first transaction (discovery), while the habit lies in the second transaction (returning to Amazon or Walmart's own apps).

A company that intends to establish habit - moat metrics will quickly discover this problem because it will focus on the ratio of customers who, after using your product to complete a task, complete a similar task again within the natural repeat cycle. For food delivery, this window is a few days. For travel, it's a few weeks. For professional tasks, it's a few months. If your repeat completion rate within this cycle is far below the industry benchmark, it means you haven't cultivated user habits; it's just a one - time use.

This changes the operational logic across industries