AI in Consumer Market: What's Missing Compared to the Internet Era?
In recent years, many have begun discussing the AI bubble, often comparing the current AI boom to the dot-com bubble: capital influx, product explosion, narrative-first, lagging implementation. AI’s efficiency gains in production have been validated across many fields, especially in programming. The skepticism mainly stems from the lack of an exponential monetization model in the consumer (C-end) space.
The reality is, with current AI technology development, the consumer market is no longer about whether features can be built, but whether long-term, high-frequency demand can be formed.
From the perspective of reach and distribution, AI doesn’t lack channels. It can fully leverage the existing internet infrastructure: apps, social platforms, recommendation feeds, search, content distribution, payments, and fulfillment—all the infrastructure is already there. AI’s problem isn’t about being seen by users, but why users should stay. Just look at how major tech companies went crazy spending money to acquire users during the Lunar New Year—how many of those users actually stayed and continued using any specific AI tool long-term?
Precisely because internet infrastructure is already complete, the challenges of consumer market implementation will be more concentratedly exposed in another set of more fundamental factors. I want to use the success mechanisms of the internet era as a reference to discuss what conditions I believe AI still needs to fulfill to succeed in the consumer market.
1. Network Effects
What really made the internet take off in the consumer space wasn’t how powerful a specific tool was, but that “connections between people” were scaled. Forums, Tieba, Weibo, short videos, group chats—all essentially did the same thing: bind content with relationships, allowing value to circulate within communities.
This brought an important result: products would grow organically. Users didn’t just consume content, they produced it; they didn’t just get information, they got recognition; they didn’t just use features, they spent time in communities. The internet’s strength was letting any group find their own corner to share and be understood, satisfying users’ psychological needs of “being needed” and “being noticed.”
In contrast, generative AI’s most common form in the consumer space is “personal conversations.” It’s certainly useful, but it naturally struggles to form community cycles: conversation results are often one-time, private, and can’t be沉淀 into public discussions; it’s even harder to form continuous relationship chain interactions and content asset accumulation like internet communities. So now, most people share their conversations with Doubao on short video platforms, sharing tutorials and ways to interact with AI—these aren’t based on AI products themselves, but still rely on internet platforms.
This doesn’t mean AI can’t build communities, but rather: if AI products stay in the individual interaction form of “I ask, you answer,” it’s hard to gain the strong network effects that the internet has. Consumer products without network effects often end up more like feature plugins than infrastructure.
2. Verification and Trust
When people discuss AI’s credibility, they often stay at the level of “does it hallucinate” or “is it correct.” But I think the more tricky problem in the consumer space is a more structural “distrust.”
Through years of development, users today have long-term “battle” experience with platform black boxes: recommendation mechanisms aren’t transparent, ads mix with content, traffic distribution has profit-driven motives, algorithms are like systems you can’t see but can influence you. After experiencing these years, users have very sensitive vigilance toward “black boxes” from large companies and their tools.
With the internet’s development, users have become familiar with these “routines” and “plays,” and may still have some vigilance. But generative AI’s black box will be even more hidden—AI black boxes might influence your thinking, becoming “deciding for you what you should believe.” OpenAI has already started adding ads, Qwen can help you shop—as a user, do you really believe these large company-based models don’t have their own tendencies and orientations? If you let a large platform’s AI help you shop, do you really believe it will choose cheaper products from other companies? And going forward, regarding certain content, will it engage in “pseudo-fact-checking?”
This means the threshold AI needs to cross in the consumer space isn’t just improving accuracy, but making “verifiability” part of the experience: sources and basis, traceability, correctability, rollback capability, explainability. Otherwise, it easily enters an awkward zone: users think it’s smart, but dare not hand over key decisions to it.
3. Motivation and Barriers: Efficiency Isn’t That Important for Ordinary People, and “Being Able to Ask Questions” Is Itself a Capability Divide
The internet’s success in the consumer space didn’t mainly come from “efficiency improvements,” but from more direct values: entertainment, information, social connection, recognition, and passing time. Many needs are essentially emotion and relationship-driven, not task-driven.
But generative AI’s most common narrative in the consumer space is “saving you time, making you more efficient.” This holds true in production, but in ordinary people’s daily lives, efficiency isn’t necessarily the primary goal. In some scenarios, the process itself is part of rest: browsing food delivery apps for a while, picking milk tea flavors, reading reviews, hesitating—this is also an acceptable form of “slacking time.” If AI just compresses these processes away, it’s essentially making people more instrumentalized.
A more realistic limitation lies in interaction barriers. AI requires you to express needs, pose questions, supplement context, and judge output quality. In other words, “whether you can use AI” is itself a capability divide. The result is often: heavy users get heavier, light users leave after two tries. The internet lowered barriers through “just click” or “just scroll,” while AI by default requires you to “explain clearly.”
So if AI really wants to achieve high-frequency, must-have demand in the consumer space, it may need not just stronger models, but lower-barrier participation methods: more structured entry points (options, templates, processes), stronger default guidance, letting passive users also obtain stable value, rather than requiring everyone to become someone who can write prompts.
I don’t believe AI has no opportunity in the consumer market. On the contrary, I’m inclined to believe it will happen, just that the path may be different from what many imagine. Although everyone says UI/UX will become obsolete, I actually think the transformation of interaction methods is very important right now—how to make chat forms simpler and easier to use, and even how to design interaction hardware entry points to make them more convenient for users. Making AI go from “usable” to “commonly used” requires not just being smarter and more intelligent, but embedding it silently into everyone’s daily life processes, providing more value for humanity.