Executive Overview
As artificial intelligence systems become more sophisticated and deeply embedded in daily academic life, universities face an uncomfortable and rapidly evolving reality. The question of whether academic departments possess adequate AI policies has shifted from a theoretical debate to an urgent institutional crisis. Recent empirical research reveals a startling transformation in higher education: a substantial and growing percentage of doctoral dissertations across both STEM fields and the humanities now contain text generated by artificial intelligence.
What began as a marginal tool for brainstorming or copy-editing has rapidly matured into a foundational drafting mechanism for doctoral candidates operating under intense pressure to publish, graduate, and secure increasingly scarce academic employment. Groundbreaking new data indicates that nearly a third of all STEM doctoral graduates in the United States completed dissertations featuring AI-generated writing by 2026. While the humanities—specifically philosophy—have historically resisted technological shortcuts, they are now experiencing a parallel, accelerated surge in AI utilization.
This structural shift forces a profound reckoning across academia. As the line between human scholarship and machine generation blurs, universities, graduate schools, and departmental committees must confront difficult questions. How should departments regulate AI use? Can these rules be effectively enforced? And perhaps most concerningly, what happens to the foundational development of intellectual expertise when the grueling, transformative process of writing a dissertation is outsourced to large language models? This report examines the data behind the AI dissertation wave, its implications for career trajectories, and the pressing need for proactive institutional policy.
Read Also
Detailed Chronology: From Novelty to Normalization
To understand how artificial intelligence became a silent co-author in contemporary doctoral dissertations, it is necessary to examine the rapid timeline of its adoption. The trajectory from total absence to widespread integration spans barely four years, reflecting the unprecedented velocity at which generative AI technologies have permeated global society.
Pre-2023: The Pre-Generative Era
Prior to the explosion of advanced generative AI models in late 2022, the drafting of a doctoral dissertation was universally understood as a purely human endeavor. While spell-checkers, grammar software like Grammarly, and basic reference management tools were ubiquitous, the generation of substantive prose, theoretical arguments, and critical literature reviews remained strictly organic. Comprehensive dataset analyses from researchers confirm that prior to 2023, zero measurable trace of AI-generated prose existed within the evaluated corpora of United States doctoral dissertations.
2023–2024: The Inflection Point and Early Adoption
Following the public launch of transformer-based large language models, graduate students swiftly recognized the utility of AI in streamlining the arduous writing process. By 2024, early adopters began experimenting with AI tools not merely for copy-editing or structural brainstorming, but for generating sections of running text.
In STEM fields, where dissertations often compile previously published journal articles, the integration of AI to polish or draft comprehensive review chapters accelerated. Even in traditionally resistant fields like philosophy, initial signs of AI utilization emerged. By 2024, data indicates that approximately 1.3% of philosophy dissertations filed in the U.S. contained detectable AI-generated writing—a modest figure, but a crucial baseline indicating the breach of a long-standing disciplinary barrier.
2025: The Acceleration Phase
By 2025, the normalization of AI assistance transitioned from underground experimentation to widespread practice. In philosophy, the proportion of dissertations incorporating at least 2% AI-generated text skyrocketed to 7.8%—a nearly sixfold increase in a single year.
Across STEM disciplines, the numbers were even more pronounced. Graduate students facing intense pressures regarding time-to-completion, combined with the normalization of AI usage in industry and standard research workflows, increasingly relied on LLMs to draft abstracts, introductions, and methodology sections. The technology evolved from a disruptive novelty into an administrative efficiency tool.
2026: The New Normal and Rapid Growth
The data from 2026 confirms that AI-assisted dissertation writing is no longer an anomaly; it is a systemic reality. In STEM fields, between 25% and 33% of all dissertations filed in 2026 contained AI-generated text. Meanwhile, preliminary data from incomplete 2026 cohorts in philosophy reveals that 11.1% of dissertations incorporate AI writing.
This rapid upward trajectory demonstrates that departmental guidelines, where they exist at all, have failed to stem the tide. The academic community has moved past the era of prevention and is now forced to grapple with the long-term consequences of machine-assisted scholarship.
Supporting Context & Metrics: The Empirical Data
The empirical foundation for these findings stems from rigorous academic research spearheaded by Daniel P. Gross (Duke University), Dror Shvadron (University of Toronto), and Hansen Zhang (Duke University). Their landmark working paper provides the first comprehensive empirical census of AI-generated writing across American doctoral programs.
Methodology of the STEM and Philosophy Studies
The researchers analyzed a near-census of U.S. STEM PhD graduates from 2019 through 2026. By processing abstracts and a stratified random sample of complete dissertations—accounting for variations across specific fields and academic years—they established that AI-generated abstracts serve as a highly reliable surrogate for AI writing throughout the full text of a document.
To determine whether these trends extended to the humanities, a collaborative extension of the study was conducted focusing specifically on philosophy dissertations. Funded by the Department of Philosophy at the University of South Carolina, this follow-up utilized the Pangram API (specifically version 4.0 model definitions) to score philosophy dissertation abstracts. AI writing was operationalized as documents containing at least 2% AI-generated text.
Key Quantitative Findings
- The STEM Landscape: AI writing was completely absent prior to 2023. By 2026, it was present in 29% of all analyzed STEM dissertations, showing continuous and rapid month-over-month growth.
- Demographic and Institutional Variations: AI utilization is notably more prevalent among international students originating from non-English-speaking countries (who may utilize LLMs to overcome language barriers) and among candidates enrolled in lower-ranked academic programs.
- The Philosophy Sector: Following a baseline of 1.3% in 2024, philosophy dissertations containing AI text jumped to 7.8% in 2025, and reached 11.1% in the partial 2026 cohort.
Methodological Challenges in Humanities Data Collection
Analyzing humanities dissertations presented unique hurdles that underscore the difficulty of tracking AI integration. Initial searches using ProQuest metadata and abstracts that listed "philosophy" as a department or subject overcounted philosophy dissertations by approximately 300%. The broad usage of the word "philosophy" in general academic discourse inflated raw counts.
When researchers restricted parameters strictly to departments explicitly named "Philosophy," new anomalies arose. Multi-disciplinary departments (such as combined "Psychology & Philosophy" administrative units) occasionally conflated degree programs, while administrative discrepancies in ProQuest records led to bizarre undercounts—such as major research institutions appearing to produce only a single philosophy dissertation across a decade. While these edge cases mean the dataset is not a complete 100% census, the directional trend remains starkly conclusive: AI writing in philosophy is accelerating rapidly.
Employment Outcomes: Productivity vs. Expertise
Perhaps the most alarming revelation of the Gross, Shvadron, and Zhang study is not merely that students are using AI to write their dissertations, but how doing so correlates with their subsequent career paths.
Diverging Career Trajectories
When linking graduates to their employment outcomes within the same program and cohort, the data reveals a clear divergence: graduates whose dissertations contain AI writing are significantly less likely to enter academic research careers. Instead, they are markedly more likely to secure non-tenure-track academic roles or transition directly into industry positions. This pattern is particularly pronounced among U.S.-origin students.
The Displacement of Expertise
This correlation cuts to the core of the academic enterprise. While artificial intelligence undoubtedly increases short-term productivity—allowing candidates to generate pages of polished prose in a fraction of the time—it appears to displace a vital component of professional formation.
Writing a dissertation is not merely an administrative hurdle or a product to be delivered; it is the primary mechanism through which intellectual expertise, deep critical thinking, and scholarly voice are forged. The struggle to articulate complex, novel philosophical arguments or synthesize intricate scientific data forces the researcher to confront the limits of their own understanding. When that struggle is outsourced to a large language model, the product may look like a dissertation, but the transformation of the student into an independent scholar may remain incomplete.
Official Statements and Institutional Dilemmas
As the academic community digests these findings, department chairs, deans, and graduate faculty are scrambling to formulate coherent responses. However, institutional approaches remain deeply fractured.
The Fine-Grained Policy Dilemma
Academic committees attempting to draft AI policies quickly encounter a philosophical and practical quagmire. Should a policy completely ban AI, or should it permit fine-grained distinctions based on the type of use?
- Is using an LLM to clean up syntax or check grammar permissible?
- Is utilizing AI to write basic python scripts for data analysis acceptable, while drafting the literature review is banned?
- Can an AI be used legitimately to summarize source material, provided the synthesis and argument are strictly human-authored?
Even if departments manage to draft policies that successfully navigate these nuanced boundaries, enforcement remains a monumental challenge. Current AI detection tools, while improving (such as the Pangram API utilized in these studies), face legal, ethical, and technical limitations when applied in formal disciplinary hearings or academic integrity investigations.
The Administrative Pressure Cooker
Compounding the internal debate over scholarly integrity is external pressure from university administration. Deans and financial officers focused on throughput, time-to-degree metrics, and institutional efficiency view the landscape through a different lens.
If a graduate student can leverage artificial intelligence to complete a dissertation in months—or even weeks—rather than years, number-crunching administrators may soon begin questioning why departments are funding multi-year residencies for writing tasks that machines can accelerate. The temptation for administrators to implicitly encourage AI adoption to reduce institutional overhead and improve graduation statistics poses a direct threat to traditional academic norms.
Future Outlook: The Need for Proactive Governance
The dissemination of empirical data regarding AI-assisted dissertations marks the end of willful ignorance in higher education. Universities can no longer pretend that generative AI is a minor distraction confined to undergraduate term papers; it has penetrated the highest levels of graduate scholarship.
The Danger of Passive Normalization
Academic norms are notoriously difficult to establish once broken. If departments adopt a passive stance—waiting to see how the trend develops—unrestricted or unmonitored AI use by graduate candidates will inevitably become normalized. Once normalization occurs, rolling back expectations will be virtually impossible. Students who choose to write entirely through human intellectual struggle may find themselves at an unfair disadvantage against peers who leverage AI to maximize output.
Recommendations for Departmental Leadership
To navigate this turbulent transition, academic departments must take immediate, proactive steps:
- Explicit Policy Formulation: Departments must move beyond vague academic honesty codes and establish transparent, specific guidelines regarding what constitutes permissible versus impermissible AI use at every stage of dissertation research.
- Redefining the Dissertation: Faculty must explicitly articulate why the writing process matters. If writing is valued as an essential tool for developing scholarly judgment, departments must defend that pedagogical value against administrative pressures for mere speed and output.
- Continuous Empirical Monitoring: Institutions should support ongoing research into how technological adoption impacts long-term scholarly output, teaching efficacy, and career sustainability.
The future of academic integrity and the preservation of deep intellectual expertise hang in the balance. The choices made by departments today will determine whether the doctoral dissertation remains a testament to human original thought or becomes merely another standardized commodity produced in partnership with artificial intelligence.
Comments
Facebook App ID not configured. Please add it in the Customizer.