I have been in the online literature circle for three years. This is the first time that editor Huahua has seen five or six AI submissions appear in her mailbox in a day: "This year, AI manuscripts have become rampant."

In the studio editor group, the latest topic of discussion has also become "How to distinguish AI authors and AI manuscripts". Some online editors collectively call these manuscripts "The DS taste is too strong" - exquisite metaphors, a faint sense of human-computer, and a sudden turn. In addition to receiving and signing the contract, online editors began to post "Boycott AI manuscripts" on social media, calling for a pure land in the online literature circle.

Internet editors dislike "AI manuscripts" Source: Screenshot of Xiaohongshu

At the same time, since the beginning of the year, online literature platforms including Tomato, Jinjiang, LOFTER and other online literature platforms have successively issued judgments or punishments for "AI manuscripts". While receiving support from some online literature authors, many novice authors who use AI to assist writing have felt panic and anxiety. They do not know what the boundaries of using AI writing and what the criteria for platform judgment are.

"I just used AI to appropriately expand and polish the content, but the traffic dropped sharply during the verification period." The novice author Qiqi told the "Silicon-based Laboratory". Qiqi said that many authors have also been "inadvertently injured by the platform and softly blocked". Some online writers began to worry that the "AI storm" affects their own profits, so they decided to wait and see for a while and open a new book after this period. (Note: The "verification period" is the recommendation mechanism of the Tomato Platform. When the author's work is updated to 80,000 words after signing the contract, he can enter the traffic pool to gain benefits. The platform provides a 7-day verification period. After 7 days, the work will be updated to 100,000 words, and the official recommendation period will enter, which is the so-called debut.)

This is not the first "confrontation" between AI and the online literature circle.

As early as July last year, "The authors jointly boycott the Tomato Novel AI Agreement" became a hot topic on social media. Over the past year, as the forefront of AIGC's implementation, the degree of integration between online literature platforms and AI has been accelerating.

From the platform, author and editor, the confusion and confusion of the entire online literature circle can be seen as a cruel and beautiful AI reality story - it seems to be an efficient productivity tool, but in the process of specific practice, when a half-maturity technology meets reality, it often faces multi-party games.

Unlike the previous round of online writers boycotting AI, in this round, the editors of online writers were lifting the table.

Since December last year, Huahua has noticed that there are more and more AI manuscripts in the email: "Sometimes you can receive five or six articles a day, and there are even many second-hand dealers who submit it repeatedly." There are generally three ways for online editors to identify "whether it is an AI draft": one is to identify AI using AI, such as using related AI detection tools such as Zhuque, Doubao, Chinaz AI, MitataAI, etc.; the second is to assist in identification. Editors will identify the main line of content, writing style, personality and specific sentences based on their own experience; the third is workThe office itself will also take corresponding measures, and will summarize the types of AI manuscripts, and blacklist relevant AI authors.

But despite the various forms of judgment, editors spend more time communicating with the author. Generally speaking, for suspected AI manuscripts, editors will communicate with the author first to avoid accidentally hurting the author. Online authors can provide self-certification - such as coded videos, document history records, etc. "We usually do not have a one-size-fits-all approach, and will communicate with the author first." Huahua said.

But in the face of the question of "what is an AI manuscript", it is difficult for editors to give an accurate answer, "The software is biased, and I am also learning how to distinguish it myself." Some online editors mentioned that after DeepSeek became popular, it promoted the AI ​​style to be gorgeous and rich, and AI became more and more aware of online writing routines, which also brought difficulty to editors to identify. During the communication process with the author, they even needed to details specific sentences and paragraphs: "For example, looking at the transition and character design is unnatural." Huahua said.

Qimao edited Lizhi gave more detailed criteria for judgment. AI was randomly pieced together. She found that a manuscript used "Hell is too cold, come and sacrifice me for me." This sentence actually comes from a classic Jinjiang work in the early years. She repeatedly confirmed with the author whether it was written by herself, and the author said "Confirm". In terms of "self-certification", Huahua mentioned that the author can provide chapter records, historical time, etc. for coded writing, and even use previous manuscripts to compare the style of the editor.

But the above method of distinguishing AI has made it difficult for some new online writers to understand. Qiqi mentioned that many online writers have AI polishing functions, and some authors have also taken their manuscripts to the detection platform. They found that the probability of AI generation is very high, which puts them in a situation of "unable to prove themselves".

A new author also posted a complaint, and the readers said that "the taste of AI is too strong", but she obviously did not use AI, "You can say that I wrote badly, but don't say that my article looks like AI."

So, the current online literature industry has fallen into a "AI vicious circle" - the platform encourages authors to use AI to assist writing, and the authors boldly try to use AI, but they have to use AI to detect and prove that they are not AI, while online literature editors have added new workloads due to AI literature.

Confuses the editors and authors of the online literature circle and the platform's attitude towards AI.

From the content scale of some online literature platforms alone, the magnitude of AI-generated content has increased exponentially. According to online big data, on March 1, 2024, Tomato's debut new books reached 5,606, 400 in the same period last year, a year-on-year increase of 1,302%. The Zhihu "Yanyan Novel" team, the top short story online literature platform, also stated in an interview with "Red Star News" that the number of AI creation content submissions has indeed increased significantly in the recent past.

The online literature platform is the first to embrace the AI ​​wave. As early as July 2023, China Literature Group launched the industry's first big model "Yiwen Miaobi" and the writer's assistant Miaobi version. Zhihu and Wall-facing Intelligent developed the "Zhihaitu AI" big model., for both the supply and demand sides of online literature, AI has been integrated into the entire process of production and distribution to experience.

After DeepSeek became popular, China Literature also announced that its author-assistant creative product, the Writer Assistant, has integrated the independently deployed DeepSeek-R1. China Literature said it will greatly enhance the writer Assistant's question-answer reasoning ability and description and polishing effect. China Online's creative model "Chinese Xiaoyao" also announced the deployment of DeepSeek. Taking Tomato Writer Assistant as an example, enter the book opening page and enter ideas. AI can provide creators with multiple inspiration references. During the specific writing process, the platform also provides an AI toolbox - including AI expansion, AI polishing, AI sequel, AI sequel, AI assistant, etc.

AI function of Tomato Writer Assistant Picture source: app screenshot

The positive attitude of the online literature platform towards AI is not difficult to understand - one is to reduce costs and increase efficiency, and the other is to optimize the business model.

From the cost of online literature platforms, the cost expenditure of online literature platforms can be divided into content costs, platform distribution costs, and film and television production costs (TV series, online dramas, movies, animations, etc.), among which the main content costs are related to creators, involving writers' remuneration, copyright procurement, etc. Taking China Literature as an example, in the first half of 2024, the proportion of content costs in revenue was 20.4%, which means that China Literature will pay nearly one-quarter of the revenue to the author.

Yiwen Cost Item Source: Financial Report

As for AI, AI will not only effectively improve the efficiency of creators, but also expand production capacity. This is undoubtedly a win-win situation for Chinawenwen.

In addition, AI can also explore more monetization methods to open up new space in member payment and IP copyright operation. For example, Tomato Novel can create a closed loop of "online literature + short drama"; China Literature's advantages lie in IP pricing and operation, and can use AI to feed back new tracks such as "IP+games, IP+going overseas".

But as Huahua felt, the platform's attitude towards AI seems to be positive, but when it comes to online creators, it is "watching while walking": "The platform's idea is to embrace and rectify." Some online editors also mentioned that the creators' attitude towards AI in this round has actually been magnified: "The platform has banned accounts that produce meaningless content and malicious hydrology in batches. In fact, there have always been accounts, but now everyone is too sensitive to AI, so it may be magnified." An online editor explained to us.

The Jinjiang Literature City, which is regarded as "the first shot in online literature to fight AI" by the online literature circle, is not completely boycotting AI. In a post on February 17, Bing Xin, the webmaster of Jinjiang Literature City, mentioned that AI writing assistance is still a new thing, academic discussion is insufficient, legal provisions are incomplete, and originality is difficult to define. "Whether AI is used" is a question that is far more difficult than originality definition.

Jinjiang Literature City issued the "Trial Operation Announcement on the Use and Judgment of AI-assisted Writing"

Source: Screenshot of Bishui Jiangting Website

This is also the current AI website articleThe biggest dilemma: First, when AI intervenes in writing, how to define what the originality of a work is? Second, how do the platform, author and editor judge "AI blot", that is, whether the author uses AI?

The current judgment criteria given by the platform focus on solving the latter problem. The platform will establish a "grading system" for AI writing assistance based on its own experience and establish a red line.

For example, Jinjiang is divided into text-type assistance and creative assistance. The text-type is divided into proofreading level, description level, and narrative level according to different degrees. Creative types are divided into element level, rough outline level, and fine outline level. Jinjiang currently accepts three levels: text-type assistance-proofing level, creative assistance-element level, creative assistance-broken assistance-broken assistance, using AI to proofread text, auxiliary outline generation elements, etc. These are actually allowed by the platform.

"The wheel of history comes faster than I thought." Bing Xin's exclamation has resonated with many people. For the online literature circle, AI is the general trend, and this shock will not be the last time. Many editors believe that the most important thing at this stage is that the platform will launch corresponding AI original definition standards in a targeted manner, and the industry will jointly establish a red line. At the same time, creators also need to adjust their mentality: "At present, pure AI literature is still easy to identify." Huahua repeatedly emphasized to us that some creators don’t have to panic too much.

She felt that in this round of AI discussion, some creators and editors stood opposite each other, which is what editors least want to see: "The online literature circle has been difficult and sufficient in recent years. We (editors) hope that the positive development of the online literature ecosystem will be more than anyone else."

Technology development comes first, platform and editors react later, and this natural lag in technology has also contributed to a group of people who "sell shovels" behind it - those who assisted AI online articles and those who anti-AI have become popular.

The AI ​​startup Waveform Intelligence, which previously launched the AI-assisted creation tool "Frog Writing", was acquired by OPPO last year. A company called GPTZero, founded in January 2023, reached millions of dollars in revenue in one and a half years by "verifying AI content".

We have learned that in addition to chatbots supported by big models, common AI tools in the AI ​​online literature circle currently have some vertical applications, such as Xingyue Writing, Mali Writing, etc.

Related AI writing tools Source: App screenshot

Search for "AI online literature" writing on social media, and many service providers have emerged, some selling online classes, and some selling software. We contacted a service provider that claimed to be "the first self-developed one-click novel vertical model for one-click in China". The profit model is to charge membership fees by word count and time - 30 days (100,000 words) is 98 yuan, 1 year (1 million words) is 560 yuan, and 1 year (unlimited words) is 2,690 yuan.

The service provider tells us that the current AI online articles "Rules for making quick money" are: "It is recommended that you write female videos, and you can get about 300-800 yuan by earning full attendance, and you can still get differentThe platform sells scrap manuscripts. "When we asked "whether there is a risk of AI detection and banning", the service provider clearly stated: "You'd better not contact the editor to sign a contract." According to him, some platforms that collect short stories currently adopt automated review, and the probability of manual review is small, so it is easier to sign a contract and pass the manuscript.

According to data released by the Chinese Academy of Social Sciences, China's online literature industry ushered in a market of 300 billion yuan in 2023, with the scale of online literature creators close to 25 million, and the number of online literature users reached 537 million. Among the 10 Chinese people on average, 4 read online literature. This is of course a vast market for AI gold digging.

In addition, the unique differentiated ecosystem of the online literature circle also determines that AI entrepreneurship is profitable.

First of all, online literary creators have large basic bases and differentiated needs, which means that AI online literary creators can quickly iterate functions according to needs, find PMFs, and quickly verify them.

Secondly, he has a strong willingness to pay. For some novice and mid-level creators, making money by using online submissions is the king. Therefore, entrepreneurs choose the right tool, coupled with the Internet's "fission" gameplay, they can soon accumulate the next group of loyal users in the private domain.

Wan Lei, the head of frog writing and co-founder of Waveform Intelligence, previously shared in an interview with "Silicon Star Man" that for novelists, writing is a tool for making a living. They are unwilling to disclose their ways to make money, and even hope that the fewer people know, the better. Therefore, "hardening products well with private domains" is also a good path.

Finally, it is the "people" among entrepreneurs. According to our understanding, in the AI ​​online literature entrepreneurship, there are a group of "great gods" with rich experience in online literature writing. They are very familiar with the workflow of online literature creation (such as popular online literature styles, book dismantling, rough outlines, detailed outlines, expansion, etc.). To a certain extent, the online literature itself has more than 20 years of commercialization, and the relevant skills, routines and frameworks have been deposited into "standard products" and "templates", which are easy to be AI-based.

As Huahua observed, in this round of AI storm, most of the people who use AI to write in batches are novice novices and AI writing studios. Their characteristic is to make money quickly and leave, and they don’t care about “raising pen names”. This is also the inevitable direction of the “Matthew effect” in the online literature circle - the top author eats meat, and the middle waist and bottom authors drink soup. As Bing Xin wrote in the forum, the AI ​​storm will not affect the top creators: “The income of top entertainment authors has not decreased but increased because their works are “anti-AI replication.” ”

The split ecology also determines the current attitude of the online literature circle towards AI differentiation. A group of technical optimists have begun to try to write with AI. In order to cope with the so-called "AI flavor", "reference for de-AI flavor" has evolved; there are also a group of conservatives who insist on "hand rubbing" and resist AI drafts, which is a phenomenon that will inevitably occur in every technological iteration.

In an online writing group, the future of "AI writing" was still discussed late at night, and optimists said: "The tools of confrontation are meaningless." A creator sent a spiritTorture: "Will online articles be meaningless in the future?"