Old Version
Life

Check Yourself

AI detection is changing how college students write papers, but unreliable scores are leaving many scrambling to meet their schools’ standards, while a growing industry promises to make their work AI-proof

By Zhao Zhouqiao Updated Nov.1

An AI-generated image of a robot checking an essay for AI-assisted plagiarism

Chen Hao, an undergraduate majoring in electronic information engineering at Huazhong University of Science and Technology in Wuhan, Hubei Province, developed an AI-generated content (AIGC) detection system as his graduation project. 

A few months ago, Chen used the system to check a classmate's dissertation. The AIGC rate was around 40 percent, well above the school's 20-percent permissible limit. 

Though the system was accurate, his classmate paid for a second check on the China National Knowledge Infrastructure (CNKI), the country's largest academic database platform, just to make sure. It came back at 41.7 percent. 

Ironically, instead of revising the paper himself, Chen's classmate turned to ChatGPT, which helped reorganize the paragraphs flagged as AIGC. This brought down his AIGC rates enough to pass official review and graduate. 

Chen's classmate is not the only student to exploit loopholes in AI detection. As AI tools become more widespread, increases in AI-assisted plagiarism in graduation theses has raised growing concerns. 

To control the use of AI tools, universities in China have set maximum AIGC rates for graduation theses. For example, Sichuan University caps the rate at 20 percent for liberal arts majors and 15 percent for STEM and medical fields. At Nanjing University of Aeronautics and Astronautics, the line is 40 percent. 

"Students who depend totally on AIGC to complete papers are skipping key steps of data collection, thought organization and logical argumentation in their research," Huang Jun, an associate professor at South-Central Minzu University in Wuhan, wrote for China Youth Daily in October 2025. 

But the argument has failed to convince some students who, like Chen's classmate, have successfully passed AI plagiarism checks. This has triggered a debate among educators over how to evaluate graduates' performance in the AI era.

Write and Wrong
Universities have reached consensus that AIGC detection is necessary to safeguard the originality of graduation theses, although their rules on AI use differ. 

Beijing Technology and Business University requires students to follow rules when using AI. Key arguments and essential data analysis should not be AI generated, while all sections involving AIGC should be identified in footnotes along with reasons why it was used and the name of the tool. 

However, rules like these are too general to address the increasingly sophisticated use of AI. 

Zhao Fang, a first-year postgraduate in English translation, told NewsChina that AIGC detection troubled her when she prepared for her paper review last year. Although she used AI only for a few grammar and expression checks in her Chinese-language dissertation, the resulting AIGC rate was 40 percent. 

All the sections flagged as AIGC were passages she claimed to have written herself, arguing that their wording and structure are the result of her studies. But to pass the review, she had to break up her sentences and reduce her use of four-character idioms. 

AI detection has confused not only university graduates but also their advisors. 

Zhang Ying, a social security lecturer at a high-ranking university, began supervising graduation theses this year. She soon realized that many set phrases and expressions in her field, such as "Basic Pension Insurance for Urban and Rural Residents," were getting flagged as AI-generated. 

The AIGC rate for the same paper can also vary across Chinese-language academic platforms, including CNKI, Wanfang Data and VIP Database, commercial products that use different training data, algorithms and recognition mechanisms. 

According to Liu Ping, an engineering major at Sichuan University, the AIGC rate for her graduation thesis was just 2 percent on PaperPass, an online Chinese-language plagiarism detection system, but 60 percent on rival platform PaperYY. Her university requires an AIGC rate below 15 percent. 

"As graduation approaches, a paper's AIGC rate is the student's new focus," Zhao told NewsChina. "Conversations are filled with inquiries about how much time and money it takes to cut the rate." 

Liu told NewsChina that a major part of her thesis was written by ChatGPT. To avoid detection, she prompted it to write in spoken language. She then used DeepSeek to revise flagged passages. Many of her ideas for prompts came from posts on lifestyle app RedNote, she said. 

Though AIGC detection rules are necessary, they should not be too general as acceptable levels can differ by discipline, Liu Haifeng, a senior professor of liberal arts at Zhejiang University in Hangzhou, told NewsChina. 

Shanghai's Fudan University prohibited AI use in research proposals and approaches in 2024. Meanwhile, at Beijing Normal University, the AIGC limit is 20 percent and students are required to mark these sections in red.

Paid Content
Although some universities provide two free AIGC checks for graduation theses through CNKI, many students prefer to pay for additional checks before submitting final versions. 

Liu spent 70 yuan (US$10) to check her nearly 40,000-character graduation thesis on CNKI. She said this was cheap compared with what some of her classmates paid, between 100 yuan (US$15) and 200 yuan (US$30) per paper. 

E-commerce platforms like Taobao and RedNote also offer ways to help students lower their AIGC rates. On PaperPass, it costs from 3-5 yuan (US$0.50-0.75) per 1,000 Chinese characters. Smaller sites, including SpeedAI, ibiling.cn and aigcleaner.com, have entered the market with even lower prices. 

However, many of these platforms funnel customers to private contacts on chat apps like WeChat and QQ where the actual transactions are negotiated. Vendors often upsell or charge extra for longer papers or AIGC rates around 15 percent. 

"The paid services are no different from what the author can do themselves to lower the rate," Zhao said. "These changes are nothing but keeping sentences short and making expressions more colloquial." 

Some of Zhang's students have paid for AIGC reduction services. While she warns that doing so violates academic ethics, she understands their dilemma. "With so much attention paid to reducing the AIGC rate, how can they concentrate on developing new ideas in their papers?" 

Some students agree. "Having to revise sentences highlighted by a detection system for four to five days straight to avoid being flagged for AI plagiarism, I don't have much time to improve the quality of my dissertation," Liu Feng, an undergraduate, told National Business Daily on June 10, 2026.

Score to Settle
Many are losing faith in AIGC detection as a litmus for authentic writing, including Chen. 

Through his research, he found that most tools identify similarities through statistical probabilities rather than track down real sources to verify originality. He now sees his own detection system more as an auxiliary tool than a reliable detector. 

Wang Wei, a computer science major at Tsinghua University, told NewsChina that unlike image differentiation, where AI generation can be distinguished from human creations through microscopic discrepancies, text detection lacks a comparable basis. 

So far, he said, such systems can only make judgments by analyzing an article's style, tone and structure, all of which can easily be imitated or revised. 

In January 2023, OpenAI launched its Text Classifier to identify AI-generated content. The program was withdrawn about six months later because of its poor accuracy, according to the company's website. 

Today, there are dozens of third-party platforms in the market, such as Originality.ai, Copyleaks and GPTZero, which was created by Beijing-born developer Edward Tian while he was studying computer science and journalism at Princeton University. 

Zhao Bin, a life sciences professor at Fudan University, began submitting his own theses and AI-generated texts to GPTZero and other detectors in May 2023. Three years on, he said AIGC detectors can hardly distinguish between AI-generated ghostwriting and writing where AI is used as an auxiliary tool. 

"So far, there are no clear metrics for AIGC detection. What the detector presents to the students is nothing but a cold number," he said. 

In an interview with Beijing Radio and Television on July 19, Huang Lei, an associate professor at the School of Artificial Intelligence at Beihang University in Beijing, noted that AI-generated and human writing styles are becoming increasingly similar and difficult to distinguish because of a new feedback loop: People are now learning to write from AI. 

But Yue Wen, director of the Department of Creative Studies at the Chinese Writers Association, disagrees, likening AI's curated sentences and rounded arguments to a delicately carved wax figure in a shop window. "It's flawless, but without the touch of life," she wrote in an article for newspaper Study Times in January.

Paper Chase
Most schools broadly agree that AIGC rates in papers should be controlled, but their approaches differ. 

The China University of Petroleum in Qingdao, Shandong Province, announced this year that academic advisors should evaluate high AIGC rated papers with more scrutiny, and thesis defense committees should also "pay particular attention" to their authors. 

At Southwest University in Chongqing, AIGC detection results serve only as a reference. Students who fail the initial review still have opportunities to resubmit their papers based on their advisors' suggestions. 

In an interview with China Central Television on May 21, Li Xin, deputy director of academic affairs at Nanjing University of Aeronautics and Astronautics, said the university's 40-percent AIGC cap was intended to guide students toward appropriate use of large language models and was not the only criterion for paper review. 

Zhang Ying said she does not oppose her students using AI while writing their papers, but AI must not replace students' own efforts in the core parts of their dissertations. 

In his dissertation published in periodical Chongqing Higher Education Research in December 2025, Wang Jiarui from the School of Law at Shandong University, argued that higher education institutions should treat AIGC rates produced by third-party organizations with caution. 

Detection should be human-centered, he said, with the thesis defense committee having the final say when evaluating. 

Wang Hongcai, a professor at the Institute of Education at Xiamen University, told NewsChina that the debate surrounding AIGC detection actually reflects shortcomings in paper-based theses. According to him, evaluations of graduates should be diversified rather than rely almost entirely on writing papers.

Degree Paths
Implemented on January 1, 2025, China's Academic Degrees Law for the first time gives practical achievements the same legal status as dissertations when awarding degrees. Now a dissertation does not have to be the only credential used to determine whether a student qualifies for a degree. 

As of May this year, creative writing graduates at Shanghai's East China Normal University received their degrees for the first time based on literary works discussed in a workshop organized by the degree defense committee. 

However, Liu Haifeng is more cautious about diversifying assessments. Unlike STEM subjects, where student graduation projects can often be used for evaluation, liberal arts subjects cannot be taught entirely without thesis training, he said. 

Learning how to write literary reviews, organize article structures and develop logical arguments are still indispensable skills for these students in these fields, Liu added. 

Some observers see these change in education as just a sign of the times. 

Most existing approaches to logical education developed during the Industrial Age may no longer be adequate for the AI era, where individualized education is becoming increasingly important, Zhao Bin argued. 

While encouraging his students to write with assistance from AI, he suggests paying greater attention to the learning process, including how students perceive, question and innovate upon established ideas. 

"It's you, though with AI assistance, who navigates the writing to its unique ending," the professor said.

Print