Does Your University Require an AI Rate Under X%? What Chinese Undergraduate, Master's, and PhD Theses Actually Require

In current public research and policy records, no universal national passing score or statutory AI percentage threshold has been established for graduation theses.

If you see neat comparison tables across search engines or social media claiming that "undergraduates cannot exceed 30%, master's theses cannot exceed 20%, and doctoral dissertations cannot exceed 10%," note that this table is not a national rule. For master's and doctoral theses, public institutional records show no verified universal numerical cutoff. School-wide graduate numbers do exist at some institutions and differ by school (for example, Lanzhou University of Finance and Economics graduate school notices trial rules from 2026-01-01 of "博士学位论文：≤15%", "全日制硕士学位论文：≤20%", and "非全日制硕士学位论文：≤25%" with theses exceeding standards requiring revision and a second check before review and defence; Guangdong University of Finance and Economics graduate school on 2025-04-01 requires a warning and revision for master's theses with an AIGC result "超过40%"); graduate students must consult the current directives issued by their own institution and department.

In practice, Chinese universities establish their own policies regarding Artificial Intelligence Generated Content (AIGC), and some delegate the specific criteria directly to individual secondary colleges. Currently documented public examples are limited in number and almost entirely derived from undergraduate thesis guidelines. Published figures differ noticeably from one institution to another, ranging from 15%, 20%, 25%, 35%, to 40%. Because this research did not establish a comprehensive national inventory across all Chinese universities, applying an unverified blanket number across different universities, disciplines, and academic years is not valid.

Understanding institutional requirements for AI thesis screening requires reviewing publicly accessible news investigations and administrative announcements issued by Academic Affairs departments. For an introductory comparison table covering four Chinese universities alongside international benchmarks, see our guide on what AI percentage is acceptable; the sections below examine how specific Chinese institutional notices frame these numbers in practice. It must be noted at the outset: the documented examples are limited in number, reflect predominantly undergraduate requirements, and do not represent a comprehensive national inventory.

The following documented cases represent specific institutional practices identified in public records:

1. Differentiating thresholds by academic discipline: Sichuan University

According to investigative reporting by Beijing Daily (2026-07-19, Thesis AI Detection Is Pushing College Students to the Brink) alongside administrative notices from Sichuan University's Academic Affairs Office (2026-04-08, citing 川大教〔2025〕28号), the institution set distinct requirements across academic fields for undergraduate theses:

• Humanities and Social Sciences: AI-generated content ratio in principle must not exceed 20%;
• STEM and Medicine: AI-generated content ratio in principle must not exceed 15%.

The same notice specifies that each individual college may independently select its AIGC screening tool ("学院可以自行选择AIGC检测工具").

1. Delegating standards to secondary colleges: Nanjing Tech (Nanjing University of Technology)

According to the same Beijing Daily report (2026-07-19), Nanjing Tech did not establish a uniform campus-wide ceiling at the central administrative level. Instead, the university authorized its secondary colleges to formulate specific implementation measures based on their respective disciplinary characteristics. Consequently, students in different departments within the same university operate under different evaluation criteria.

1. Universities establishing ceilings or pass/fail lines at 40%: TUST, NUAA, Hebei Engineering, and Guangxi Normal
2. Tianjin University of Science and Technology (TUST): The Academic Affairs Office notice (2024-04-09) states that "AIGC检测结果显示的智能生成内容比例原则上不超过40%（不含40%）" (the proportion of AI-generated content in principle must not exceed 40%, excluding 40%). Above this ratio, students receive a warning, revise, and undergo re-testing; colleges may formulate more specific professional requirements on top of the campus guideline ("更具体、更专业的要求").
3. Nanjing University of Aeronautics and Astronautics (NUAA): The Academic Affairs notice for the 2026 undergraduate cohort stipulates that "AIGC比例高于40%或未按规定时限提交AIGC检测的，检测结果视为不通过" (an AIGC ratio exceeding 40% or failure to submit testing on time is deemed a failing result). Students who fail cannot take the first-round defense, which is postponed by at least one month. Reports from Beijing Daily and Xinhua also describe this as a university-wide rule that the AIGC proportion must not exceed 40% ("规定AIGC比例不得高于百分之四十").
4. Hebei Engineering University: Beijing Daily (2026-07-19) reported that the university sets an undergraduate AIGC ceiling of "不得高于百分之四十" (must not exceed 40%); the institution's internal notice dated 2026-04-24 is restricted to campus network access and cannot be viewed externally.
5. Guangxi Normal University: Both Beijing Daily (2026-07-19) and Xinhua reported that Guangxi Normal enforces a university-wide rule that the AIGC proportion "不得高于百分之四十" (must not exceed 40%).

1. Setting a campus-wide 35% pass line: Aba Teachers University

Academic Affairs notices at Aba Teachers University establish a uniform campus-wide passing threshold: a notice on AI tool usage dated 2025-03-21 states "AIGC检测率≤35%，则视为通过" (an AIGC rate ≤ 35% is deemed passing), while the 2026-03-31 notice for the 2026 cohort specifies "AIGC小于35% … 则检测通过，可参加毕业答辩；如检测结果大于规定标准，则检测不通过" (AIGC under 35% passes and qualifies for graduation defense; results exceeding the standard fail).

Misleading comparison tables and the rules governing tool use Comparing these documented cases demonstrates that published undergraduate and graduate thresholds differ across institutions, ranging from 15%, 20%, 25%, 35%, to 40%, while other institutions leave the standard to departmental discretion. The widely circulated internet table claiming "30% for bachelors, 20% for masters, 10% for PhDs" is not an established national rule; while tiered graduate numbers do exist at a few schools (such as 15%/20%/25% at Lanzhou University of Finance and Economics), standards vary widely across institutions, and readers should never treat any single online formula as a universal regulation.

Whether students may use AI-assisted rewriting, humanizer, or editing tools depends entirely on the academic integrity regulations of their specific institution, department, and assignment. While some institutions permit language polishing under explicit disclosure rules, others strictly restrict any generative assistance. Readers should consult their own institutional guidelines rather than assuming all tools are either automatically allowed or universally prohibited, and must never rely on commercial marketing promising guaranteed detector outcomes.

Much of the anxiety surrounding screening scores occurs because students treat an automated percentage as a definitive institutional verdict. In academic administration, these represent three distinct, progressive stages:

• Stage 1 (Automated Screening): Software scans the text and outputs an automated statistical percentage;
• Stage 2 (Warning and Departmental Review): If the score crosses an institutional threshold, it triggers an advisory alert, faculty review, or mandatory revision and re-testing; consequences differ significantly by school (ranging from required revisions to defense postponement), and candidates must follow their own institution's notice;
• Stage 3 (Substantive Misconduct Inquiry): Where suspected plagiarism, fabrication, unauthorized ghostwriting, or serious integrity violations arise, relevant university or departmental academic committees initiate formal review and substantive evaluation.

1. Automated Screening Threshold (机检初筛阈值)

This is the initial score generated by an automated scanning platform (such as CNKI or VIP) after scanning a manuscript.

• Nature: It represents an algorithmic estimate based on statistical text patterns, serving as an automated preliminary filter.
• Limitations: Automated tools cannot examine an author's actual writing or research process. Highly standardized terminology or formal citation structures can produce elevated percentages. Therefore, an initial screening score is only an administrative reference point.

1. Warning Threshold and Departmental Review (预警整改线)

When a preliminary score exceeds an institution's published guideline, it initiates an administrative review process rather than an immediate disciplinary decision.

• Standard Protocol: The student and faculty advisor receive an automated advisory notification.
• Remediation: Specific institutional consequences vary by school. Some institutions require revision followed by a mandatory re-check (such as TUST and Aba Teachers University), some treat an exceeding score as an outright failing result that postpones defense by at least a month (such as NUAA), while others require revisions to meet standards before defense or blind review. Students must consult their department's directives to confirm the required remediation, whether submitting draft records, revising flagged text, or completing formal re-testing.

1. Formal Misconduct Finding (学术不端认定)

Academic misconduct is a formal disciplinary determination governed by institutional integrity regulations. For a detailed examination of why an automated flag does not equal an integrity violation under university statutes, see whether a high AI rate constitutes academic misconduct.

• Institutional Authority: An automated detection score provides an algorithmic reference, while actual evaluation authority rests with faculty supervisors, secondary college academic committees, or institutional review bodies.
• Human Inquiry: Final disposition of flagged theses depends on institutional procedures (for instance, TUST delegates final determination of results to secondary college academic committees); verifying actual research artifacts (such as thesis proposals, incremental draft histories, laboratory notebooks, and empirical datasets) remains the core evidentiary basis for resolving integrity questions.

Students and faculty frequently observe significant score variations across different tools and subject areas:

1. Detection platforms use different algorithms and reference bases

CNKI, VIP, Wanfang, and international systems rely on distinct classifier algorithms and reference databases. The exact same thesis file submitted to different commercial tools will frequently yield divergent percentages. Consequently, running unauthorized pre-checks on commercial websites does not provide a reliable indicator of an institution's official test result.

1. Inherent differences in disciplinary writing styles

This structural difference explains why universities such as Sichuan University implement distinct thresholds by academic field:

• Humanities and Social Sciences: Manuscripts rely on extended argumentative narratives, theoretical synthesis, and contextual literature analysis.
• STEM and Medicine: Theses require standardized experimental protocols, universal terminology, mathematical formulas, and structured technical descriptions. Because scientific reporting conventions naturally rely on standardized phrasing, applying a single blanket algorithm across all disciplines risks producing distortions, leading some institutions to adopt tiered or departmental standards.

Rather than relying on informal forum discussions, degree candidates should systematically verify their programme's official requirements using the following 5-step checklist:

• [ ] 1. Review the current year's directive from your Academic Affairs Office or Graduate School Consult your university's official portal for the current academic cycle's graduation project guidelines. Do not rely on outdated notices from previous cohorts.
• [ ] 2. Check departmental and secondary college implementation rules Confirm with your departmental administrator or counselor whether your specific college has issued detailed criteria tailored to your major.
• [ ] 3. Confirm the official platform designated by your institution or college Identify whether the university or your specific secondary college uses CNKI, VIP, Wanfang, or another tool (noting that some institutions, such as Sichuan University, permit individual colleges to select their own AIGC detection tools), and submit exclusively through approved channels rather than third-party services.
• [ ] 4. Clarify the review and remediation procedure if a warning threshold is crossed Determine the exact institutional protocol: How many resubmissions are permitted, and what written explanation or process documentation must be submitted to the thesis advisor?
• [ ] 5. Maintain verifiable evidence of your drafting and research process Preserve time-stamped draft files, tracked editorial revisions, literature notes, raw laboratory datasets, and survey records throughout the research cycle. These original artifacts serve as the definitive substantive proof of independent authorship.

The journalistic reporting and institutional notices cited in this article include the following records (Note: some institutional notices may be archived on departmental pages or require campus network credentials; readers are advised to search institutional portals by notice title):

1. Beijing Daily (北京日报客户端): "论文AI检测，正在‘逼疯’这届大学生" (Thesis AI Detection Is Pushing College Students to the Brink). Investigative report published 2026-07-19 noting Sichuan University's discipline-specific thresholds (humanities ≤ 20%, STEM/medicine ≤ 15%), Nanjing Tech's departmental delegation, and university-wide guidelines at NUAA, Hebei Engineering, and Guangxi Normal University ("规定AIGC比例不得高于百分之四十"). URL: `https://xinwen.bjd.com.cn/content/s6a5ccb7de4b03fa51a81b89b.html`
2. Academic Affairs Office, Tianjin University of Science and Technology (天津科技大学教务处): Notice on undergraduate thesis similarity and AIGC screening (published 2024-04-09, stipulating that AI-generated content in principle must not exceed 40%, excluding 40%, with warnings, revision, and re-checking for scores exceeding this threshold). URL: `https://jw.tust.edu.cn/ggtz/65b256ce743a41fa82c3533e16009807.html`
3. Academic Affairs Office, Nanjing University of Aeronautics and Astronautics (南京航空航天大学教务处): "关于做好2026届本科毕业设计（论文）后期工作的通知" (Notice on later-stage undergraduate thesis work for the 2026 cohort, published 2026-04-16, stipulating that an AIGC ratio exceeding 40% or failure to submit on time is deemed failing and bars students from the first-round defense). URL: `https://aao.nuaa.edu.cn/2026/0415/c8230a396991/page.htm`
4. Academic Affairs Office, Hebei Engineering University (河北工程大学教务处): Campus administrative notice (internal notice dated 2026-04-24 restricted to campus network; 40% threshold confirmed by media reporting). URL: `https://jiaowu.hebeu.edu.cn/`
5. Academic Affairs Office, Aba Teachers University (阿坝师范学院教务处): "阿坝师范学院本科生毕业论文规范使用AI工具的通知" (Notice on standardizing AI tool usage in undergraduate theses, 2025-03-21, setting AIGC rate ≤ 35% as passing) and "阿坝师范学院教务处关于启动2026届本科生毕业论文（设计）学术不端、形式规范检测和AIGC检测工作的通知" (Notice on launching academic misconduct, format, and AIGC screening for the 2026 cohort, 2026-03-31, stipulating that AIGC below 35% passes and qualifies for graduation defense). URL: `https://jw.abtu.edu.cn/info/1100/4592.htm` / `https://jw.abtu.edu.cn/info/1100/5094.htm`
6. Academic Affairs Office, Sichuan University (四川大学教务处): "关于开展2026届本科毕业论文（设计）学术不端行为检测工作的通知" (Notice on academic misconduct screening for 2026 undergraduate theses, published 2026-04-08, citing 川大教〔2025〕28号, setting humanities ≤ 20% and STEM/medicine ≤ 15%, and authorizing individual colleges to select their AIGC detection tools). URL: `https://jwc.scu.edu.cn/info/1069/10518.htm`
7. Science and Technology Daily / The Paper (科技日报 / 澎湃新闻): "毕业论文将检测AIGC率，检测标准科学吗？" (Graduation Theses to Test AIGC Rate: Are Detection Standards Scientifically Sound?). Investigative report published 2026-05-22 examining algorithmic boundaries and institutional reflections. URL: `https://m.thepaper.cn/newsDetail_forward_33210288`
8. China National Radio (央广网): "代降论文AI率服务乱象调查：学生遭遇欺诈还被威胁" (Investigation into Deceptive 'AI Rate Lowering' Services: Students Defrauded and Threatened). Investigative report published 2025-07-14 (originally reported by Legal Daily / 法治日报) documenting fraud, poor-quality rewrites, and extortion threats to report students' theses to universities. URL: `https://news.cnr.cn/native/gd/20250714/t20250714_527260647.shtml`
9. Graduate School, Lanzhou University of Finance and Economics (兰州财经大学研究生院): "关于进一步加强研究生学位论文学术不端行为检测工作的通知" (Notice on further strengthening academic misconduct detection for graduate theses, trial rules effective from 2026-01-01, establishing thresholds of doctoral dissertations ≤ 15%, full-time master's theses ≤ 20%, and part-time master's theses ≤ 25%, with theses exceeding standards requiring revision and a second check before review and defense). URL: `https://yjsy.lzufe.edu.cn/__local/6/0E/C9/E637BE6416B028CC2A39386BB10_4741E860_2A3BE.pdf`
10. Graduate School, Guangdong University of Finance and Economics (广东财经大学研究生院): "关于开展2025届夏季毕业研究生学位论文AIGC检测工作的通知" (粤财大研〔2025〕11号, published 2025-04-01, issuing warnings and requiring revisions for master's theses with AIGC scores exceeding 40%). URL: `https://yjsc.gdufe.edu.cn/2025/0401/c3011a217797/page.htm`