AI-driven self-media operators are swarming to churn out accounts in batches, and the banks of the fish pond are packed with fishing enthusiasts.

Over the past two years, the most buzzing spot in the tech industry is not large model launch events, but a batch of vertical AI accounts that have suddenly sprung up in the self-media track. Countless people have transformed to run AI-focused self-media accounts. Can they succeed in this transformation?

01 The Transformation of Real Estate Self-Media Practitioners

Last winter, I ran into Lao Zhou at a café in Wangjing.

Lao Zhou used to run real estate-focused accounts, and he could not stop talking about "land auctions", "premium rate" and "deep waters of regulation". That day, a Word document was spread out on his laptop, with a striking title: Post-90s Doctor Dropped Out to Start a Business, Reconstructing the Trillion-Dollar Track with Large Models. The first line of the main text reads: "At 3 a.m. in Beijing, the parameters of large models are still changing."

I asked Lao Zhou, "Aren't you writing about the property market anymore?"

Lao Zhou lit a cigarette, with the excitement of finding a new oil field in his eyes: "Are there any fish left in the property market? Now all the fish across China are in the AI fishpond. Do you know QbitAI? As a vertical account, it has reaped huge benefits from this AI boom. One of my peers, who used to write about pitfalls to avoid in catering franchise last year, renamed his account 'XX AI Observation' this year, and gained 80,000 followers in three months. 80,000 followers! I have been writing about the property market for five years, and my number of followers is not even half of his."

He lowered his voice, as if passing on the coordinates of a good fishing spot: "The trick is super simple. Find an AI company that has raised financing, go through its official website, press releases, and past interviews with the founder, feed all the materials to the large model to get a story framework, then wrap it up with financial narrative techniques — top talents leaving big factories, giving up millions of annual salary, with a dream of AGI. Make the title striking enough: He Gave Up Millions of Annual Salary Just to Let Machines Understand Human Hearts. The traffic will pour in nonstop."

I poured cold water on him: "Why would companies let you write about them?"

Lao Zhou smiled: "The problem now is not that companies refuse to accept interviews, but that they are too busy to receive so many media."

02 Perspective from an Entrepreneur

I originally thought that was just drunk talk. Until last month, I met A K, an entrepreneur working on Agent direction.

A K's company closed its angel round last year. He told me that on the day the financing news was officially announced, he witnessed a spectacle in the history of self-media.

"In just half a day, dozens of articles about our team popped up across the whole network." A K's expression was a mix of absurdity and exhaustion. "The first article was written in a very emotional tone, saying that I had 'lain low for five years, and buried myself in tackling general artificial intelligence in the basement'. The more I read, the more distracted I got — my apartment doesn't even have a basement. It also said that I 'resisted capital doubts and went through countless sleepless nights'. To be honest, I slept very well during that period."

He clicked on the second and third articles, and found that each story version had its own "innovation": some said he focused on medical AI, some said he was deeply engaged in autonomous driving. But in essence, his company develops small models for code assistance.

"I reached out to the author to make corrections. The other party was very polite, saying that the article was generated by AI and manually fine-tuned, with materials collected from public networks, and it was impossible to verify every single point. The article had already been published, and the reading volume was rising rapidly, so it was unrealistic to withdraw it." A K spread his hands, "Listen to that — 'it's unrealistic to withdraw the article'. Now even rumor-mongering is done with such 'professional ethics', only making up half of the facts."

What's even more outrageous was at an industry exhibition. There were dozens of exhibitors on site, and before the meeting was over, a large number of "exclusive interviews" had appeared online. Dialogue with XX Founder: How We Break Through in the Second Half of the Computing Power Track. A K came across several so-called exclusive interviews with himself, and the "original remarks" attributed to him were things he had never said. He sent a private message to the editor, who was very straightforward: we need to produce more than a dozen articles a day, and it is completely unrealistic to conduct offline interviews one by one, and everyone is doing it this way.

"Exclusive interview," A K repeated the term, "the meaning of exclusive is — I am the only one who has never met this interviewer."

03 Fast-Produced Articles in the AI Era

This is unimaginable in the old days of information dissemination.

In the past, the process of tech reporting was like this: find clues, arrange interviews, verify data, cross-check information, and spend a whole week polishing a deep-dive article. It was slow, but there were solid anchors. Readers could tell which part was fact and which part was judgment.

Now this assembly line is completely reversed: priority is given to output, update frequency, and eye-catching titles. Facts have become raw materials in the material library, and stories have become the shell to wrap traffic. What is AI best at? Filling in information gaps. If the materials you feed it are incomplete, it will not leave a blank, but make up plots that are more attractive than reality. Non-existent basements, investment institutions that have never rejected projects, government-enterprise orders that have not been landed. When readers scroll through the information feed, how can they tell which part is from a real interview and which part is the reasonable imagination of the model?

They can't. And ordinary readers are exactly the group least capable of distinguishing — they have no channels to verify the contracts, indicators and team background of an AI company. When they see ten articles all saying that this enterprise has unlimited prospects, they will naturally take it as the industry consensus. When it is later exposed that the product fails to meet expectations and the financing market cools down, people suddenly realize that those overwhelming startup stories before are mostly castles in the air with no foundation.

As a result, enterprises fall into the absurd situation of "being unable to defend themselves even with a hundred mouths": when they try to clarify, the clarification article is drowned in the massive content, and the more they explain, the more it looks like a sophistry. A founder told me that he is most afraid of the official announcement of financing news now. In the past, he was afraid of having no money, but now he is afraid of having money — half a day after the announcement, the whole network is holding a feast, and he himself becomes the dish on the table.

04 The Fishpond Theory Comes into Effect

Of course, some people will say: Didn't those leading accounts like QbitAI also rise by interpreting information in batches?

This statement is only half right. The early dividend of AI content was built on the information gap. In 2024 and 2025, a large number of practitioners could not understand overseas papers or keep up with model iterations. Those who were willing to do translation, sorting and popularization of knowledge naturally had value. At that time, the supply was scarce, readers were hungry for content, and enterprises were also happy to see more industry voices. Besides, behind those leading accounts are real editorial teams who have long been deeply rooted in the industry, with AI only serving as an auxiliary tool, not the main production subject.

Later imitators completely reversed the order: AI changed from an auxiliary tool to the main production subject, and humans only need to polish and publish the articles. Once one account works well, it will be scaled up into a matrix immediately — the boss registers three to five entities, Account A writes highlights of startups, Account B does industry reviews, and Account C specifically compares parameters of different companies. I know a young content entrepreneur who registered eight AI accounts at one go, hired two people to run prompts and modify titles, and stably produced more than 20 articles a day. The update volume is fully maximized, creating a scene of prosperity.

The fishpond theory starts to take effect at this step.

In a water area that seems to have plenty of fish, the first one or two anglers can easily get a good harvest. When the news spreads, hundreds of people flock here with fishing rods, standing densely on the bank with intertwined fishing rods and flying bait. What's worse, they are using the same brand of bait. A hundred accounts use similar prompts, grab the same batch of news, and generate articles with similar styles. Today Account A writes "dark horse breaking through the encirclement", tomorrow Account B writes "disrupting the industry", the day after tomorrow Account C rewrites the inspirational story again. The content is getting more and more similar, and the only difference is the wording of the titles.

Fish have sharp senses. Continuous disturbance on the water surface and movements from all directions will make them instinctively alert and sink into deep water.

The first to perceive the change are enterprises. In the early years, a reliable in-depth article could help the team recruit talents, connect with investors and reach customers, so PR budgets were willing to tilt to tech content. Now when many founders see AI accounts coming to them, their first reaction is vigilance: will this article make up information, exaggerate technology, and fabricate dialogues? If there is a mistake after publication, the cost of refuting rumors is extremely high, and it may even mislead investors and customers. As a result, many companies quietly tighten the supply of materials, and some even set rules: no interviews and no materials will be provided to any self-media that generates content in batches with AI. It is equivalent to the fish actively swimming to a quieter water area, no longer approaching the groups of anglers on the bank.

Then it comes to readers. After seeing too many identical startup success stories, people have figured out the routine: as long as the title is "genius team, cutting-edge technology, about to disrupt the world", they will first mark it with a question mark. In the past, people discussed technology in the comment section of touching stories, but now the top comments under similar articles have become: "This article must be written by AI right?" "It's the same old script again." The appeal of such content fades, and the traffic level drops accordingly. The same title routine could easily go viral in 2025, but in the second half of 2026, the data declines visibly. Platform algorithms are also identifying highly homogeneous AI-generated content, and quietly reducing its weight.

That entrepreneur running the account matrix cut half of his accounts half a year later. He calculated an account for me: the unit price of traffic has dropped, enterprise PR teams are getting more and more cautious, and the price of commercial orders has been squeezed to the floor. An AI-generated article seems low in cost, but when tens of thousands of similar articles across the network compete together, it loses its scarcity. "I thought I had found a gold mine, but after I came in, I found that the gold mine was already crowded with gold diggers, covered in dust, and no one could see the direction clearly."

05 Self-Media Practitioners Also Need to Build Core Competitiveness

Lao Zhou came to see me again recently. This time he didn't talk about the myth of gaining followers, but said something that impressed me a lot.

"Maybe I have to learn to dive," he flicked the cigarette ash, "There are too many anglers, and the fish won't bite the hook anymore. But there are real valuable things under the water — people who truly understand technology, people who are really building products, people who have gone through multiple cycles. They don't bite the hook, but you can swim down and stay with them."

Let me translate this sentence, it is actually the common sense of this industry: AI is good at summarization, laying foundation work and drafting, but the core of industry reporting will always be human observation, face-to-face dialogues, and independent judgment after multi-party cross-verification. When writing about AI enterprises, some people are willing to stay on the front line to see the real pain points in landing, and record the parts that will never appear in those success stories — failures, disputes, capital pressure, and the tedious trial and error day after day. Admitting that many startup projects have no earth-shattering reversal, only the perseverance that cuts slowly like a blunt knife. This kind of content that can stand verification is the real scarce product.

Those accounts that only churn out content in batches relying on templates and imagination will sooner or later hit the wall of the red ocean. They thought that mastering production tools means mastering the track, but they forgot the most underlying logic of communication: trust cannot be consumed repeatedly. If you cheat the reader once, he will scold you; if you cheat him ten times, he will stop reading — not only stop reading this one account, but also stop reading all content in this whole pond.

06 Epilogue

Writing these is not to sing down the AI content industry. This is the standard script of every boom: the early adopters find a wild pond with lush water and plenty of fish, then latecomers flock in with fishing rods, surround the fishpond tightly, and then either leave disappointedly with their rods, or those who really know how to swim take off their vests and dive into the water.

The success of those leading accounts like QbitAI is written in the deep water, not in the sound of swinging fishing rods.

Therefore, all new accounts that are gearing up by the AI fishpond should think clearly about one question: are you the one who can catch big fish, or the noise that scares the fish away?

After all, in a field that seems to have plenty of fish, when there are too many anglers, the fish will move away. And when the fish move, they never say hello — they just stop biting the hook one day, leaving a group of people on the bank reviewing whether their titles are striking enough facing empty buckets.

When the tide recedes, the ones left on the bank will not be the players who set up accounts the fastest and update the most, but only the few who are willing to wait patiently, do not disturb the water surface excessively, and take pains to distinguish the true from the false.

What is truly scarce in this track has never been one smooth and beautiful startup story after another, but real, restrained, and verifiable industry observations.

This article is from WeChat Official Account "Kaisen Observation", author: Kaisen, published with authorization from 36Kr.