AI-like writing appears in 29% of U.S. STEM dissertations. What does that mean?
STEM
The fields of science, technology, engineering, and mathematics.
AI detector
A tool that estimates whether text may have been generated by AI.
working paper
A research paper shared before formal peer review.
What happened
A new NBER working paper examined U.S. STEM dissertations from 2019 through 2026. STEM means science, technology, engineering, and mathematics. Daniel P. Gross, Dror Shvadron, and Hansen Zhang found AI-generated writing in 29.4% of dissertations filed from January through May 2026. The rate was 1.7% in 2023, 5.5% in 2024, and 18.8% in 2025. Chemical & Engineering News reported the findings.
The wording matters. The study did not show that 29.4% of dissertations were written by AI. It found writing that an AI detector judged likely AI-generated. A dissertation might contain a small edited passage, a translation, or much larger AI-written sections. The percentage alone cannot separate these cases.
What the number measures
Researchers examined STEM dissertations submitted by U.S. schools and placed in publicly available thesis and dissertation repositories. The study used an AI detection tool to estimate whether sections looked machine-generated. It measured detected writing, not every kind of AI assistance. It also did not prove that every flagged passage was actually written by AI.
Use was uneven across fields and student backgrounds. In the 2026 group, AI-like writing appeared in 36% of mathematics and computer science dissertations, 35% of engineering dissertations, and 22% of physical science dissertations. It appeared in 41% of dissertations from students in countries where English was not the main language. The figure was 23% for students from mainly English-speaking countries.
Why it matters
A PhD dissertation is more than a final document. Writing forces students to choose evidence, explain limits, and revise weak ideas. Those steps can help build scientific expertise. Generative AI can also help students translate or improve English. That support may be especially useful for researchers who write in a second language.
The paper also reports a career pattern. Within the same program and cohort, graduates whose dissertations contained AI writing were less likely to enter academic research. They were more likely to enter industry or non-tenure-track jobs. The paper presents this as an association. It does not prove that AI caused those career outcomes. Students who use AI may differ in other ways.
What is confirmed
The clearest result is the rapid rise in detected AI-like writing in U.S. STEM dissertations. The result does not describe all students, all countries, or all academic papers. The study is also a working paper. It has not yet completed formal peer review.
What remains uncertain
The detector may not reveal how a student used AI. It may not distinguish translation, proofreading, brainstorming, and full drafting. Accuracy may also change as language models and editing tools change. Most importantly, the study cannot yet show whether AI improves research quality or weakens long-term training. The career relationship needs more evidence before it can support a cause-and-effect claim.
What to watch next
Universities may need clearer rules for different kinds of AI help. They may ask students to disclose their use and explain their choices aloud. Some experts also suggest placing more weight on oral examinations. That would test how students think, develop a study, and defend its results.
The central question is not simply whether students use AI. It is whether they still do the difficult thinking that research training is meant to teach.
Nearly three in ten U.S. STEM dissertations show AI-like writing
📰 Full story: AI-like writing appears in 29% of U.S. STEM dissertations. What does that mean?
A study found a sharp rise in AI-like text inside science dissertations.
dissertation
A long paper about a student’s research.
AI detector
A tool that guesses whether writing looks AI-made.
STEM
Science, technology, engineering, and mathematics.
💡 The gist
- 29.4% of U.S. STEM dissertations showed AI-like writing in early 2026.
- The share rose from 1.7% in 2023.
- The study found a pattern, not proof that AI caused harm.
A dissertation is a long paper about a student’s research. STEM means science, technology, engineering, and mathematics. Researchers checked dissertations from U.S. schools and public online collections.
They used an AI detector. The tool guessed whether some writing looked machine-made. It did not watch students write. It also could not show how much AI help each student used.
The numbers grew quickly. The share was 5.5% in 2024. It reached 18.8% in 2025. It reached 29.4% from January through May 2026. Mathematics and computer science had the highest share. Their figure was 36%. Engineering followed at 35%. Physical science was 22%.
The study also found different rates by language background. The share was 41% for students from countries where English was not the main language. It was 23% for students from mainly English-speaking countries. AI may help with translation and English editing. That can make scientific writing easier.
But writing is also practice. Students learn to explain evidence. They learn to notice weak ideas. They learn by revising their work. Researchers worry that too much AI help could remove part of that practice.
The study found another pattern. Students with AI-like writing in their dissertations were less likely to enter academic research. They were more likely to enter industry or other research jobs. This does not prove AI caused their career choices. Other differences may explain the result.
The study is a working paper. It has not completed formal peer review. Its detector may miss some AI writing. It may also flag writing that was only edited. More research must test writing quality, student learning, and future careers.
Schools may ask students to explain their research aloud. They may also ask students to report their AI use. The goal is to check thinking, not only polished writing.
Reference: the research paper and the news report.
A computer helper appeared in many big school papers
📰 Full story: AI-like writing appears in 29% of U.S. STEM dissertations. What does that mean?
People found more computer-made words in long research papers.
dissertation
A very long paper about research.
AI detector
A tool that guesses whether AI made words.
A dissertation is a very long school paper. Grown-ups write one after much research.
AI is a computer tool that helps with words. People looked at U.S. STEM dissertations. In 2026, 29.4% contained AI-like writing. The number was much smaller in 2023.
This does not mean AI wrote every paper. An AI detector makes a careful guess. A guess is not proof.
Some students may use AI to fix English. That can make writing easier. People still need to practice thinking. Writing helps people explain their ideas.
Researchers do not yet know the long-term result. Schools may ask students to explain their work aloud. That helps show what they understand.
Reference: the research paper.