Can you spot the bot? Identifying AI-generated writing in college essays
Open accessAbstract
The release of ChatGPT in 2022 has generated extensive speculation about how Artificial Intelligence (AI) will impact the capacity of institutions for higher learning to achieve their central missions of promoting learning and certifying knowledge. Our main questions were whether people could identify AI-generated text and whether factors such as expertise or confidence would predict this ability. The present research provides empirical data to inform these speculations through an assessment given to a convenience sample of 140 college instructors and 145 college students (Study 1) as well as to ChatGPT itself (Study 2). The assessment was administered in an online survey and included an AI Identification Test which presented pairs of essays: In each case, one was written by a college student during an in-class exam and the other was generated by ChatGPT. Analyses with binomial tests and linear modeling suggested that the AI Identification Test was challenging: On average, instructors were able to guess which one was written by ChatGPT only 70% of the time (compared to 60% for students and 63% for ChatGPT). Neither experience with ChatGPT nor content expertise improved performance. Even people who were confident in their abilities struggled with the test. ChatGPT responses reflected much more confidence than human participants despite performing just as poorly. ChatGPT responses on an AI Attitude Assessment measure were similar to those reported by instructors and students except that ChatGPT rated several AI uses more favorably and indicated substantially more optimism about the positive educational benefits of AI. The findings highlight challenges for scholars and practitioners to consider as they navigate the integration of AI in education.
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
Heavy dependence on AI detection tools may lead to a surveillance-oriented approach to education, ultimately eroding trust between instructors and students.
“Heavy dependence on AI detection tools may lead to a surveillance-oriented approach to education, ultimately eroding trust between instructors and students [14].”
From Evaluating the Effectiveness and Ethical Implications of AI Detection Tools in Higher Education (2025) · cites this paper as [14]
In “Theoretical contributions and practical implications” (page unavailable), the authors caution against relying on AI detection tools because they are imperfect and false accusations can cause serious harm. The paper does not say that reliance on these tools creates a surveillance-oriented approach or erodes trust between instructors and students. (LLM, conf=0.84)
Experimental — accuracy not yet measuredStudents' real-world experiences, especially their struggles with unclear institutional guidelines on AI usage, remain underexplored and insufficiently documented.
“Students' real-world experiences, especially their struggles with unclear institutional guidelines on AI usage, remain underexplored and insufficiently documented [14].”
From Evaluating the Effectiveness and Ethical Implications of AI Detection Tools in Higher Education (2025) · cites this paper as [14]
In “Attitudes toward AI in education” (page unavailable), the paper reports that some students wanted guidance on whether uses such as converting a document to MLA format would violate academic integrity. In “How AI-generated content poses challenges for educational assessment” (page unavailable), it says there is little empirical evidence about ChatGPT’s impact on education and best practices. It does not specifically establish that students’ real-world experiences with unclear institutional guidelines are underexplored or insufficiently documented. (LLM, conf=0.84)
Experimental — accuracy not yet measuredAI detection tools are inherently flawed and risk unjustly penalizing students due to false positives.
“They argue that such tools are inherently flawed and risk unjustly penalizing students due to false positives.”
From Evaluating the Effectiveness and Ethical Implications of AI Detection Tools in Higher Education (2025) · cites this paper as [14]
Another study shows that even experienced instructors struggle to distinguish between AI- and student-written work, underscoring the need for rethinking how learning is evaluated.
“Similarly, another study highlights the vulnerability of traditional essay-based assessments, showing that even experienced instructors struggle to distinguish between AI-and student-written work, thus underscoring the need for rethinking how learning is evaluated [14].”
From Evaluating the Effectiveness and Ethical Implications of AI Detection Tools in Higher Education (2025) · cites this paper as [14]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
The citation lists haven’t been fetched yet
References 45 · Cited by 49
Research works printed but not linked: 32
- Al Darayseh A (2023) Acceptance of artificial intelligence in teaching science: Science teachers' perspective. Computers Educa- tion: Artif Intell 4:100132. https://doi.org/10.1016/j.caeai.2023.100132 · DOI
- Bertram Gallant T (2011) Creating the ethical academy. Routledge, New York
- Biswas SS (2023) Potential use of Chat GPT in global warming. Ann Biomed Eng 51:1126-1127. https://doi.org/10.1007/ s10439-023-03171-8 · DOI
- Borenstein J, Howard A (2021) Emerging challenges in AI and the need for AI ethics education. AI Ethics 1:61-65. https://doi. org/10.1007/s43681-020-00002-7 · DOI · Paper page
- Bretag T (ed) (2016) Handbook of academic integrity. Springer
- Bretag T, Harper R, Burton M, Ellis C, Newton P, Rozenberg P, van Haeringen K (2019) Contract cheating: a survey of Australian university students. Stud High Educ 44(11):1837-1856. https://doi.org/10.1080/03075079.2018.1462788 · DOI · Paper page
- Brown TB, Mann B, Ryder N, Subbiah M, Kaplan JD, Dhariwal P, Neelakantan A, Shyam P, Sastry G, Askell A, Agarwal S, Herbert- Voss A, Krueger G, Henighan T, Child R, Ramesh A, Ziegler D, Wu J, Winter C, Amodei D (2020) Language models are few- shot learners. Adv Neural Inf Process Syst 33. https://doi.org/10.48550/arxiv.2005.14165 · DOI · Paper page
- Chi MTH, Wylie R (2014) The ICAP framework: linking cognitive engagement to active learning outcomes. Educational Psychol 49(4):219-243. https://doi.org/10.1080/00461520.2014.965823 · DOI · Paper page
- Cizek GJ (1999) Cheating on tests: how to do it, detect it, and prevent it. Routledge R Core Team (2021) R: A language and environment for statistical computing R Foundation for Statistical Computing, Vienna, Austria. https://www.
- Curtis GJ, Clare J (2017) How prevalent is contract cheating and to what extent are students repeat offenders? J Acad Ethics 15:115-124. https://doi.org/10.1007/s10805-017-9278-x · DOI
- Devlin J, Chang M-W, Lee K, Toutanova K (2018) BERT: pre-training of deep bidirectional transformers for language understand- ing. ArXiv. https://doi.org/10.48550/arxiv.1810.04805 · DOI · Paper page
- Dwivedi YK, Kshetri N, Hughes L, Slade EL, Jeyaraj A, Kar AK, Baabdullah AM, Koohang A, Raghavan V, Ahuja M, Albanna H, Albashrawi MA, Al-Busaidi AS, Balakrishnan J, Barlette Y, Basu S, Bose I, Brooks L, Buhalis D, Wright R (2023) So what if Chat- GPT wrote it? Multidisciplinary perspectives on opportunities, challenges, and implications of generative conversational AI for research, practice, and policy. Int J Inf Manag 71:102642. https://doi.org/10.1016/j.ijinfomgt.2023.102642 · DOI · Paper page
- Eke DO (2023) ChatGPT and the rise of generative AI: threat to academic integrity? J Responsible Technol 13:100060. https:// doi.org/10.1016/j.jrt.2023.100060 · DOI · Paper page
- Erickson S, Heit E (2015) Metacognition and confidence: comparing math to other academic subjects. Front Psychol 6:742. https://doi.org/10.3389/fpsyg.2015.00742 · DOI
- Fischer I, Budescu DV (2005) When do those who know more also know more about how much they know? The development of confidence and performance in categorical decision tasks. Organ Behav Hum Decis Process 98:39-53. https://doi. org/10.1016/j.obhdp.2005.04.003 · DOI
- Fleming SM, Weil RS, Nagy Z, Dolan RJ, Rees G (2010) Relating introspective accuracy to individual differences in brain structure. Science 329:1541-1543. https://doi.org/10.1126/science.1191883 · DOI · Paper page
- Gigerenzer G (1991) From tools to theories: a heuristic of discovery in cognitive psychology. Psychol Rev 98:254. https://doi. org/10.1037/0033-295X.98.2.254 · DOI
- Gigerenzer G, Hoffrage U, Kleinbölting H (1991) Probabilistic mental models: a brunswikian theory of confidence. Psychol Rev 98(4):506-528. https://doi.org/10.1037/0033-295X.98.4.506 · DOI · Paper page
- Gunser VE, Gottschling S, Brucker B, Richter S, Gerjets P (2021) Can users distinguish narrative texts written by an arti- ficial intelligence writing tool from purely human text? In C. Stephanidis, M. Antona, & S. Ntoa (Eds.), HCI Interna- tional 2021, Communications in Computer and Information Science, (Vol. 1419, pp. 520-527). Springer. https://doi. org/10.1007/978-3-030-78635-9_67 · DOI · Paper page
- Jakesch M, Hancock JT, Naaman M (2023) Human heuristics for AI-generated language are flawed. Proceedings of the National Academy of Sciences, 120(11), e2208839120. https://doi.org/10.1073/pnas.2208839120 · DOI · Paper page
- Jiang F, Jiang Y, Zhi H, Dong Y, Li H, Ma S, Wang Y, Dong Q, Shen H, Wang Y (2017) Artificial intelligence in healthcare: past, pres- ent and future. Stroke Vascular Neurol 2(4):230-243. https://doi.org/10.1136/svn-2017-000101 · DOI · Paper page
- Joo YJ, Park S, Lim E (2018) Factors influencing preservice teachers' intention to use technology: TPACK, teacher self-efficacy, and technology acceptance model. J Educational Technol Soc 21(3):48-59. https://www.jstor.org/stable/26458506
- Köbis N, Mossink LD (2021) Artificial intelligence versus Maya Angelou: experimental evidence that people cannot differentiate AI-generated from human-written poetry. Comput Hum Behav 114:106553. https://doi.org/10.1016/j.chb.2020.106553 · DOI · Paper page
- Salvi F, Ribeiro MH, Gallotti R, West R (2024) On the conversational persuasiveness of large language models: a randomized controlled trial. ArXiv. https://doi.org/10.48550/arXiv.2403.14380 · DOI
- Shynkaruk JM, Thompson VA (2006) Confidence and accuracy in deductive reasoning. Mem Cognit 34(3):619-632. https://doi. org/10.3758/BF03193584 · DOI
- Susnjak T (2022) ChatGPT: The end of online exam integrity? ArXiv. https://arxiv.org/abs/2212.09292 · arXiv
- Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez AN, Polosukhin I (2017) Attention is all you need. Adv Neural Inf Process Syst 30. https://doi.org/10.48550/arxiv.1706.03762 · DOI · Paper page
- Waltzer T, Dahl A (2023) Why do students cheat? Perceptions, evaluations, and motivations. Ethics Behav 33(2):130-150. https:// doi.org/10.1080/10508422.2022.2026775 · DOI
- Waltzer T, Cox RL, Heyman GD (2023a) Testing the ability of teachers and students to differentiate between essays generated by ChatGPT and high school students. Hum Behav Emerg Technol 2023:1923981. https://doi.org/10.1155/2023/1923981 · DOI · Paper page
- Waltzer T, DeBernardi FC, Dahl A (2023b) Student and teacher views on cheating in high school: perceptions, evaluations, and decisions. J Res Adolescence 33(1):108-126. https://doi.org/10.1111/jora.12784 · DOI
- Weidinger L, Mellor J, Rauh M, Griffin C, Uesato J, Huang PS, Gabriel I (2021) Ethical and social risks of harm from language models. ArXiv. https://doi.org/10.48550/arxiv.2112.04359 · DOI · Paper page
- Wixted JT, Wells GL (2017) The relationship between eyewitness confidence and identification accuracy: a new synthesis. Psychol Sci Public Interest 18(1):10-65. https://doi.org/10.1177/1529100616686966 · DOI · Paper page
- Authors
- Waltzer, T., Pilegard, C., & Heyman, G. D.
- Affiliations
- University of California San Diego (US)
- Corresponding author
- Waltzer, Tal
- Venue
- International Journal for Educational Integrity
- Published
- 2024-07-08
- Volume
- 20
- Issue
- 1
- ISSN
- 1833-2595 (ISSN-L)
- OpenAlex
- W4400400166
- Cited by
- 49
- References
- 45
- Citations by year
- 2026: 26 · 2025: 20 · 2024: 3
- FWCI
- 18.42
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Locations
- International Journal for Educational Integrity (publishedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
- OSF Preprints (OSF Preprints) (submittedVersion)
- Funders
- National Science Foundation (NSF SPRF-FR# 2104610)
- Topics
- Physical Sciences › Computer Science › Information Systems › Artificial Intelligence in Education
- Social Sciences › Safety Research › Academic integrity and plagiarism
- Physical Sciences › Computer Science › Artificial Intelligence › Authorship Attribution and Profiling
- Keywords
- AI-generated text detection, ChatGPT, attitudes toward AI, artificial intelligence in education, content expertise, academic integrity
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
