Summary of Key Points
This news article highlights the controversy surrounding the use of AI in assessing university graduation theses: Students often stay up late to revise their theses in order to meet the minimum AI content threshold of 20%-40% (even lower for postgraduate works), repeatedly paying for checks, and sometimes even distorting the original content. The results from different assessment platforms vary significantly due to lack of transparency in the evaluation criteria, often leading to unfair judgments against genuine work. These platforms profit by offering both testing services and essay paraphrasing tools. Experts and official guidelines recommend shifting from relying solely on AI-based assessments to a more comprehensive process-oriented evaluation method that teaches students how to use AI effectively.
1. The Struggle of Students to Reduce AI Content in Their Theses
Many students go to great lengths to meet the requirements: Zhou Xiaoyun from Chongqing stayed up until dawn to reduce her thesis’s AI content percentage from 70% to 20% in order to qualify for the defense. A postgraduate student from a top-tier university in central China spent 780 yuan on seven checks, only to find that parts of her original experimental sections were mistakenly identified as containing significant AI elements. Even worse, original content is frequently misclassified—summaries, English translations, and acknowledgments can all be marked as AI-generated. To pass the assessments, students are forced to simplify technical terms (for example, replacing “wabi-sabi aesthetics” with “rough and simple”) and rearrange sentence structures, sometimes at the expense of the quality of their work. Universities only provide two free checks, so students often use multiple platforms for additional verification, incurring costs ranging from 100 to 300 yuan per check, with some spending much more.
2. Uncertain Assessment Results: The Same Thesis, Different Scores on Various Platforms
The randomness of assessment results is frustrating for students: An article generated by AI was rated as having 0% AI content by CNKI but 55.71% by VIP. The same thesis received scores that differed by more than 50% on platforms like PaperPass and PaperYY. This inconsistency stems from the varying algorithms used by different platforms, which focus on different aspects of the text (e.g., abstracts or the overall structure), as well as differences in pre-trained materials and models. Ironically, well-written, logically clear texts are more likely to be flagged as AI-generated, while poorly structured, colloquial language is sometimes considered original. Students describe this process as akin to playing the lottery, unsure of which check will result in a passing grade.
3. The Profit-Making Tactics of Assessment Platforms
AI assessment has become a lucrative business for these platforms: According to the 2025 financial report of CNKI’s parent company, Tongfang Holdings, its AIGC (Artificial Intelligence Generated Content) assessment business generated over 100 million yuan in revenue, with a 200% increase in usage. Platforms like VIP and Wanfang not only charge for assessments (CNKI charges 2 yuan per thousand characters, while VIP charges 38 yuan per document) but also offer essay paraphrasing services (10 yuan per thousand characters), sometimes providing so-called “revision tips.” The market is flooded with unofficial paraphrasing tools on e-commerce platforms, however, with no guarantee of quality. There are numerous complaints about issues such as significant discrepancies in results and misleading marketing. Industry experts criticize these platforms for being both assessors and providers of paraphrasing services, essentially acting as both “referees” and “participants” in the evaluation process.
4. A Way Forward: Focus on How Students Use AI, Not Just the Numbers
Students are not entirely opposed to AI-assisted assessments but prefer a more reasonable approach. Guidelines released in May this year suggest moving away from relying solely on technical checks towards a system that includes “transparent disclosure of methods used” (e.g., by keeping records of interactions with AI tools). Professor Zhao Bin from Fudan University has adopted this approach, avoiding direct AI content checks and instead focusing on the students’ interaction with AI during the writing process to demonstrate their thinking. He believes that rather than fearing students’ use of AI, it is better to teach them how to use it effectively, emphasizing the importance of critical thinking skills over machine-generated numbers.
This news article reveals the new challenges in academic evaluation in the age of AI: the limitations of technical assessments and the need to align with educational goals. The solution may not be to completely ban AI but to integrate it in a way that serves as a tool for students, rather than a burden.