99% of AI-Generated Content Is Junk: When the AI Tech Carnival Becomes a "Content Manure Machine

Abstract

The cliff-like drop in the technical barriers to AI has made "building AI" as easy as playing with building blocks. However, behind this technological revelry lies a frenzy of "content manure machines" — a surge of homogeneous, low-value AI products springing up like mushrooms only to wither away just as quickly. This article delves into the three major dilemmas of AI entrepreneurship: the proliferation of "half-baked products" due to plummeting technical costs, severe homogenization in vertical niches, and the trend towards "junkification" of AI-generated content. At the same time, it points the way for entrepreneurs who truly want to make a mark in the AI field: only by genuinely understanding human needs, deeply grasping industry pain points, and building a sustainable business model can one gain a firm foothold in the wave of AI entrepreneurship.

Main Body

The AI scene these past two years has been surreal. Just a couple of years ago, it felt like high-tech confined to labs. Now, anyone with a bit of computer know-how can cobble together an AI tool over a weekend.

Model API costs have been on a rollercoaster ride, plummeting from $20 per million tokens in 2023 to just a few cents now — practically free. The result is AI products flooding the streets. It looks lively, but they shut down even faster. According to Dang.ai's statistics, out of over 5,000 tools tracked, more than 1,200 have vanished without a trace. Right at the start of 2025, 193 died off, like chives being harvested, one batch after another.

1. A Record of Cemetery Expansion: Those AI Tools That Vanished Overnight

1.1 Even Star Projects Can't Hold On

Take MiniMax's "Everything Tracker." It was hyped as miraculous when it launched, claiming to track information across the entire web. Yet, it shut down within months of public testing. Why? The numbers just didn't add up. Compute costs were terrifyingly high, and a viable commercialization model never materialized. The team eventually pivoted to enterprise services. Then there's StepFun's "Bubble Duck," a role-playing AI that seemed trendy. But user acquisition stalled. Eventually, its consumer-facing business was merged into a new platform, leaving just a skeleton crew to maintain it. Xiaohongshu's Trik AI drawing tool fared even worse. Criticized by illustrators for copyright infringement, it was directly taken down, its account deleted, vanishing without a sound.

1.2 Vertical Niches Are Like a Revolving Door

Companion AIs are the most crowded niche. Copycats of Character.AI, like "Huanhuan AI" and "Paper Plane AI," initially rode the wave of traffic dividends. Once the hype faded, they all fizzled out. There's also "momoAI Emotional Companion," which announced its shutdown at the end of May, just after a major update in April. Users complained, "Just as I started to trust it, I got dumped." And let's not forget those "AI Romance" apps like AlienChat and LoveBird — all offline now. The first wave of people trying to date AIs have collectively experienced heartbreak.

2. Tech Got Cheaper, But Fake Demand Piled Up

2.1 The Cost Avalanche: Anyone Can Build a "Half-Baked Product"

How easy is it to build an AI tool now? GPT-3.5's API price was slashed to one-tenth of its original cost. OpenAI's new o3 model cut costs by another 80%. It works out cheaper than printer paper. Anthropic's Claude 3.5 Haiku costs just $0.8 per million tokens for input. The result? Grab a few scripts, wrap them in a ready-made UI component library, and you can slap together a product in three to five days. The problem is, while things get built fast, the demand is often unvalidated. Many products are born from "a flash of inspiration," without ever asking users if they actually want it.

2.2 The Skin-Swapping Game Is Unsustainable

In the writing, drawing, and companion niches, hundreds of "skin-swapped tools" popped up within six months — just changing logo colors and tweaking prompts, with essentially the same features. Users see a sea of "look-alikes" that don't solve new problems, so they naturally can't be bothered. It's like someone looking for a "meal replacement" but finding streets full of things that look like burgers but taste wrong — you take one bite and throw it away.

3. Are 99% of AI Contents "Digital Garbage"?

There's just too much AI-generated content online now. Media outlets talk about a "red tide" or a "sea of synthetic sludge," but that's putting it mildly. Some programmers call Vibe Coding's output "spaghetti code." In my view, these AI-generated contents and products are also "spaghetti," and the underlying AI models are non-stop "manure machines."

The models themselves are starting to "consume" low-quality content, generating worse and worse output, creating a vicious cycle. Users on X and Reddit complain that "99% of AI content is garbage." The Financial Times even named "Slop" its 2024 Word of the Year, specifically describing this useless information. Platforms like Medium are being forced to manually delete posts and verify content, or risk being drowned in junk.

Where's the core problem? The technical barrier is low; anyone can generate content, but no one teaches you how to judge "whether this thing has any value." What's truly scarce has never been the content itself, but the insight to "connect human needs, industry pain points, and business logic into a closed loop," and the execution power to "quickly do the right thing well." People's creativity has been ignited by AI, but many are just in it for "the thrill of creation," like posting on social media — they feel good doing it, without caring if anyone else needs it. Like a friend of mine who constantly shares his random musings in various group chats. Asked why, he says, "I thought it was interesting." He only cares about his own enjoyment, completely disregarding whether others can stand it.

4. To Do Something Meaningful with AI, Focus on These Three Points

4.1 Understand Human Nature to Capture Users

Character.AI now has over 20 million monthly active users. Why is it hot? It lets users customize their own characters for chatting, companionship, coaching, and even built a community for everyone to play together. Users can project their emotions there, like chatting with a "virtual idol," paying for subscriptions and character skins. The key is, it doesn't force-feed features but follows users' emotional needs, making the AI feel like "someone who gets you."

4.2 Dive Deep into Industry Pain Points

Why can medical AI get into hospitals? Because it genuinely helps doctors save time and reduce errors, cutting diagnosis time by 30%-50% and lowering missed diagnosis rates. Hospitals directly integrate it into standard procedures. Enterprise RAG systems can automatically organize knowledge, saving 40% on customer service manpower. These are tangible, calculable benefits. Unlike some tools that are hyped to the skies but end up as useless ornaments in practice — why would a business pay for that?

4.3 The Business Model Must Form a Closed Loop

For B2C, rely on subscriptions and selling virtual items. For B2B, charge based on usage or offer integration services. Entrepreneurial teams must do the math from the start. The user's Lifetime Value (LTV) must be higher than the Customer Acquisition Cost (CAC). Otherwise, even if compute costs are cheap, you can't fill the bottomless pit of traffic costs. I've seen a team working on AI copywriting that only focused on user acquisition. Their LTV never caught up to their CAC. Once the funding ran out, they had to shut down.

5. In Short, You Must Know What Users Actually Want

Many AI products die today not because the tech is bad, but because they fundamentally misunderstand user needs. Tech is cheap, tools are handy, but the core capability still lies with people — the ability to see through human nature, accurately identify industry pain points, and straighten out the business logic. It's like how anyone can buy a power drill now, but not everyone knows where to drill the hole. AI is just a tool, amplifying human creativity and the desire for expression. Without demand insight + a business closed loop + execution power, even the strongest model can only produce "synthetic waste" and "manure machines." Those who survive in the future will undoubtedly be the ones who can fuse "what the technology can do" with "what users need." Otherwise, even if you create something flashy with AI, you'll just end up as another name in the "cemetery."

References

Dang.ai — AI Tools Directory — The AI tools directory site referenced in the text. It catalogs 5000+ AI tools and maintains "AI Cemetery" dead-link statistics, serving as the source for the data point on "massive numbers of AI tools shutting down rapidly."
Financial Times: Year in a word — Slop — The original Financial Times article naming "slop" as the 2024 Word of the Year, specifically referring to low-quality content generated in bulk by AI.