Why Might Writing Teachers be Concerned about AI?

Daniel Libertz

Various AI programs, and the large language models (LLMs) they rely on, have become widely accessible to writers since 2022. In this chapter, I want to offer some of the reasons why your instructors might restrict, forbid, or not engage with AI in your writing class.

Let’s begin with how these programs work. In an article for researchers of writing, Ryan Omizo explains that the most contemporarsions of LLMs available for common commercial use for writing (e.g., ChatGPT, Claude, Gemini) consist of a “transformer” that takes texts that a user inputs into the program’s interface and then “predicts one token [word, part of a word, or punctuation mark] at a time based on the likelihood of that token/word occurring in the output string” (200). The transformer uses what is called “self-attention” where a series of mechanisms finds “the relevance of each term” in relation to “every other term in the sequence as a series of attention scores…[and the] transformer then uses these attention scores to ‘retrieve’ the most likely next token given the model’s training” to display in the output (200). In other words, a user inputs text, the program “scores” it based on the data it was trained on (i.e., lots of text written by humans), and, based on those scores, the program returns to the user a series of words that are likely to occur together in a sequence. This technology has now developed so far that these transformations of user text consistently become plausible-sounding writing.

As you may have experienced, teachers of writing have engaged with these programs in a variety of ways: embracing them and integrating them into their classes, allowing some limited use, or actively forbidding them. This chapter explores some common reasons for caution around AI programs in the writing classroom.

Some common concerns writing teachers have about AI are:

1. student learning
2. writing quality and homogeneity
3. bias and misinformation
4. authorship
5. ethical concerns not directly related to writing (e.g., environment, health, privacy, democracy)

Concern About Student Learning

One of the more intriguing studies about learning and writing is by Nataliya Kosmyna et al. They measured the brain activity of 54 writers who used a search engine, used an AI program, or used only their brain. They found that the group with the weakest cognitive activity was the group who used AI—that is, they had the least activation of connections along the neural pathways in their brains. Naturally, research like this could lead your writing professor to be hesitant to engage with AI in their classroom.

More specifically, many of your instructors see learning to write as connected to learning to think effectively. As Annette Vee has argued, LLMs do a great job confidently providing answers, but the space of uncertainty—a place to linger and turn over ideas in our minds—is an important space for learning, one that LLMs can shut down.

There is some research that supports Vee’s position. For instance, in a study of over 600 participants who used AI, Michael Gerlich found that those who used AI most frequently had lower critical thinking skills. Gerlich explains that this could be due to “cognitive offloading,” or the use of external aids to take on work that reduces “engagement in deep, reflective thinking” (2). While Gerlich argues for integrating AI in education together with more active learning, some writing professors may wonder what role, if any, is appropriate for AI in reading and writing at the college level. After all, much of what we ask of students requires advanced levels of critical thinking.

Concern About Writing Quality and Homogeneity

Writing that uses generative AI can feel “off” to some readers. Ryan Omizo and William Hart-Davidson note in their research of 20,216 Amazon product customer reviews that LLMs are particularly good at reproducing “genre conventions”—that is, specific featuresof a type of text that occur often like the format of a lead paragraph in a news article or academic transition phrases like “In conclusion”. However, they tend to be worse at the more human move of “genre signaling”—that is, the rhetorical collection and progression of conventions to “prim[e] responses from audiences” (2). What distinguishes genre signaling from genre conventions, is that signaling is the organization of conventions into communication that is recognizable, familiar, and engaging. Humans, unlike generative AI programs, are very good at being aware of their audiences and anticipating how they might perceive the writer’s words. Thus, humans have more variety in the way they do genre signaling. More simply, Omizo and Hart-Davidson and other researchers suggest that humans make a wider and more interesting range of rhetorical choices in their writing overall. At the sentence level, for example, John Gallagher points out the tendency of AI programs to overrely on templates, or patterns of writing. It is human to use templates, but because AI programs produce what is common, they overuse templates. Gallagher argues these programs “write rehashes of the same tired templates.” This can make AI writing both predictable and very boring.

Another set of scholars, Moon, Green, and Kushlev compared over 2,000 college admissions essays generated by human writers and by AI and found that the essays by humans are much more diverse in terms of ideas and creativity. Findings like these prompt concerns that widespread adoption of LLMs for writing may lead to less innovation, creativity, and cultural diversity in writing produced overall.

To that latter point of cultural diversity, there have also been concerns about how LLMs tend to “sound” the same, regardless of who is using them. Most LLMs are trained on a large collection of language data that primarily originates from public websites like Reddit or Wikipedia (ODSC – Open Data Science). If this writing that LLMs rely on are more likely to be written by certain groups of people, then writing outputs will start to sound like those groups of people—namely: White, from the U.S., and male. Guo, Shang, and Clavel describe it this way: “as LLMs become more prevalent in content creation, their outputs may trend towards homogenization, risking a loss of linguistic richness” (1519).

Overall, many writing teachers are concerned about LLMs having a negative influence on the kinds of writing students can produce, leading to writing that does not sound very human, is boring, diminishes possibilities for innovation in writing and thought, and/or discourages linguistic diversity.

Concern About Bias and Misinformation

In the previous section, I mentioned how the data LLMs are trained on can lead to issues with linguistic diversity. For similar reasons, many AI programs are predisposed to produce outputs that are more based on stereotypes than reality. For instance, Hofmann et al. found that when prompting chatbots to evaluate text using the dialect of African American English (AAE) vs. the dialect of Standard American English, outputs consistently contained stereotypical associations with Black people in relation to AAE texts. Your teachers may be concerned about how to properly use AI programs in class when considering risks of biased information as it pertains to marginalized groups and in other ways training data might lead to bias, whereas other information retrieval methods have a reduced risk for such outputs.

You may be familiar with AI “hallucinations” already. These occur when AI generated text sounds plausible but contains inaccuracies. Because LLMs predict text combinations, rather than verifying information through possession of knowledge, output is always at risk of being false. For a topic that’s new to you, it can be easy to miss wrong answers. Computer scientist Hassan Elsayed warns that the “persistent allure of instant answers fosters an environment of minimal questioning” making any user more vulnerable to accepting too quickly output that may be inaccurate (22). This can happen to students, but it also happens frequently to professionals like lawyers (Kaste). Though you may have some instructors who encourage you to use AI programs to find information (e.g., sources for your papers), other instructors might worry about the higher risk of misinformation compared to teaching you how to competently use a search engine, evaluatesources, and read those sources as a non-expert.

Concern About Authorship

AI companies have used questionable practices in preparing datasets for their models to train on. For instance, OpenAI “has said it would be impossible to create tools like its groundbreaking chatbot ChatGPT without access to copyrighted material” (Milmo). Without use of copyrighted materials, many of these companies would produce poorer performing models. However, the way these companies gain access to and use copyrighted materials call forth ethical questions about how writers and creators should be (and often are not) compensated, and how much agency they have to say “no” to these companies. Your instructors—many of whom are published authors themselves—may feel it sets a bad example for young writers to use products that violate the wishes (and perhaps, the rights) of so many other writers. They may also wish for you not to share the writing of other people—including other students—without knowledge of the writer’s position on using their work.

One of the duties of your writing teacher is to teach you about proper citational techniques—that is, giving proper credit to the work of other writers you use in your writing. The lack of transparency in what professional and public writers contributed to the output generated by AI programs make citation a difficult issue since their writing was used as training data which makes attribution impossible. As an example, if you go to Wikipedia about any topic, there is a very thorough accounting of references and even edits from users on the Wikipedia page itself (to see who works on and writes Wikipedia pages you can simply click “View History” to see what accounts worked on the page and the history of the page’s changes). AI programs can be prompted to include explanation of sources used, but because of how AI programs function, these outputs still require verification due to the risk of hallucinations. Therefore, your teachers may worry about you getting into bad habits and using inaccurate or made-up citations.

A final concern about authorship—perhaps one you are most familiar with—are questions of academic integrity when it comes to your own authorship. While there is not scholarly consensus on generative AI’s impact on academic integrity for college students, one systematic analysis of studies of this topic point to concerns about the ability of these programs to “quickly generate sophisticated texts, which can be misused for creating undetectable, ghostwritten assignments” (Bittle and El-Gayar 5-6). Your instructor may see how easy it is to produce text and feel that the temptation for students to be unethical is too vast to justify endorsing direct engagement with AI in your writing class. Furthermore, AI detection tools are inadequate (Bittle and El-Gayar 6) and frequently produce false positives (Cooperman and Brandão; Liu et al.; Liang et al.; D’Agostino; Kling). Therefore, some instructors may view avoidance of AI use for writing as the easiest way to avoid academic integrity violations. Instructors may view what counts as an academic integrity violation via AI in different ways further complicating engaging with AI (however, see Gallagher, Wagner, and Canzonetta for one study that tries to find some consensus and nuance here related to writing, especially infographic on page 8).

Ethical Concerns Not Directly Related To Writing

Your instructor may have concerns about encouraging use of AI in their classroom due to ethical challenges not related to writing. The one you probably hear about most relates to the environment. As scientists at NASA explain, scholarly consensus has established that humans have caused a continual warming of the earth due to primarily to the use of fossil fuels (“Scientific Consensus”). It is also true that data centers that AI companies rely on consume a lot of fossil fuels (Yañez-Barnuevo, “Data Center Buildout”). There are other concerns, too, about water usage, though these concerns are mostly related to the building of data centers in drought-prone areas (Yañez-Barnuevo, “Data Centers and Water”). While many technologies are energy-hungry and water access can be a big problem in many ways, some instructors may question whether using AI programs warrants further warming of the planet and restriction of water access.

Some instructors are also concerned about the potential mental health impacts of AI use. For instance, the phenomenon of “AI psychosis” involves users developing fixations with chatbots that amplify existing psychological issues (Wei). There have been reports of chatbot users committing acts of violence, engaging in self-harm, or even committing suicide after engaging with chatbots for a period of time (Ramsland).

Other instructors may be concerned about their own privacy or the privacy of their students. Because AI programs, in some situations (like in schoolwork), need a lot of access to user information, there is risk in terms of the kind of information provided and what happens to it after it is provided. For instance, some apps can get access to your calendar, photos, browsing history, and stored passwords on your browser (Whittaker). Allowing access to such information could lead to nefarious things done by tech companies but just their possession of your information creates opportunities for hackers to gain needless access to your personal data.

Finally, some instructors may have beliefs about democracy and concentrations of power, which could persuade them to keep AI out of their classrooms. In probably the most extreme example of negative effects of AI on democracy, xAI’s Grok has a history of being updated to better reflect the opinions of the company’s owner, Elon Musk (Horvath). Some instructors may be concerned about engaging with commonly available AI apps that can be easily updated to give further undue influence to the world’s most powerful people.

Conclusion

Like the use of any technology, there can be pros and cons. For some, the pros outweigh the cons; for others, it is the opposite. This chapter has no goal toward persuasion. Instead, I wanted to provide a recounting of five specific perspectives on AI discouragement, non-engagement, or refusal in a brief chapter. Other resources like this blog post and this website may also be helpful for exploring these perspectives (McIntyre; Sano-Franchini, McIntyre, and Fernandes) may also be informative to you. I hope this chapter can be useful as a way to start discussion on limitations of AI programs in relation to writing and its presence in your writing classroom, whether you (or your instructor) uses AI or not.

Works Cited

Bittle, Kyle, and Omar El-Gayar. “Generative AI and Academic Integrity in Higher Education: A Systematic Review and Research Agenda.” Information, vol. 16, no. 4, 2025, pp. 296-310.

Cooperman, Steven R. and Roberto A. Brandão. “AI Tools vs. AI Text: Detecting AI-Generated Writing in Foot and Ankle Surgery.” Foot & Ankle Surgery: Techniques, Reports & Cases, vol. 4, no. 1, 2024, pp. 1-5.

D’Agostino, Susan. “Turnitin’s AI Detector: Higher-Than-Expected False Positives.” Inside Higher Ed, 1 June 2023, https://www.insidehighered.com/news/quick-takes/2023/06/01/turnitins-ai-detector-higher-expected-false-positives. 17 July 2026.

Elsayed, Hassan. “The Impact of Hallucinated Information in Large Language Models on Student Learning Outcomes: A Critical Examination of Misinformation Risks in AI-Assisted Education.” Northern Reviews on Algorithmic Research, Theoretical Computation, and Complexity, vol. 9, no. 8, 2024, pp. 11-23.

Gallagher, John R. “The Growth of Template Rhetoric in an AI Society.” John Gallagher’s Technology Newsletter, 26 May 2026, https://meresophistry.substack.com/p/the-growth-of-template-rhetoric-in. 17 July 2026.

Gerlich, Michael. “AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking.” Societies, vol. 15, no. 6, 2025, pp. 1-28.

Guo, Yanzhu, Guokan Shang, and Chloé Clavel. “Benchmarking Linguistic Diversity of Large Language Models.” Transactions of the Association for Computational Linguistics, vol. 13, pp. 1507-1526.

Hoffman, Valentin et al. “AI Generates Covertly Racist Decisions About People Based on their Dialect.” Nature, vol. 633, pp. 147-154.

Horvath, Bruna. “Grok, Elon Musk’s AI Chatbot, Seems to Get Right-Wing Update.” NBC News, 7 July 2025. https://www.nbcnews.com/tech/elon-musk/grok-elon-musks-ai-chatbot-seems-get-right-wing-update-rcna217306. 17 July 2026.

Kaste, Martin. “Penalties Stack Up as AI Spreads Through the Legal System.” NPR, 3 Apr. 2026, https://www.npr.org/2026/04/03/nx-s1-5761454/penalties-stack-up-ai-spreads-through-legal-system. 17 July 2026.

Kling, James. “Prof Accused of being AI Bot.” The Exponent, 26 July 2023, https://www.purdueexponent.org/campus/article_2d1826e2-2bfa-11ee-84c9-6f34496edb29.html. 17 July 2026.

Kosmyna, Nataliya et al. “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task.” arXiv. 31 Dec. 2025. https://arxiv.org/abs/2506.08872

Liang, Weixin et al. “GPT Detectors are Biased Against Non-Native English Writers.” Patterns, vol. 4, no. 7, 2023, pp. 1-4.

Liu, Jae Q. J. et al. “The Great Detectives: Humans Versus AI Detectors in Catching Large Language Model-Generated Medical Writing.” International Journal for Educational Integrity, vol. 20, no. 8, 2024, pp. 1-14.

McIntyre, Megan. “Talking to Students about Generative AI: 3 Questions For Teachers and Students to Consider.” FEN Blog, Composition Studies, https://compstudiesjournal.com/2026/03/09/talking-to-students-about-generative-ai-3-questions-for-teachers-and-students-to-consider/. 17 July 2026.

Milmo, Dan. “’Impossible’ to Create AI Tools Like ChatGPT Without Copyrighted Material, OpenAI Says.” The Guardian, 8 Jan. 2024, https://www.theguardian.com/technology/2024/jan/08/ai-tools-chatgpt-copyrighted-material-openai. 17 July 2026.

Moon, Kibum, Adam E. Green, and Kostadin Kushlev. “Homogenizing Effect of Large Language Models (LLMs) on Creative Diversity: An Empirical Comparison of Human and ChatGPT Writing.” Computers in Human Behavior: Artificial Humans, vol. 6, 2025, pp. 1-10.

ODSC-Open Data Science. “The Top 10 LLM Training Datasets for 2026.” Open Data Science Conference, 9 Apr. 2026, https://odsc.medium.com/the-top-10-llm-training-datasets-for-2026-40578afa9f89. 17 July 2026.

Omizo, Ryan. “Research Brief: Transformers.” College Composition and Communication, vol. 77, no. 1, 2025, pp. 197-209.

Omizo, Ryan and William Hart-Davidson. “Is Genre Enough? A Theory of Genre Signaling as Generative AI Rhetoric.” Rhetoric Society Quarterly, vol. 54, no. 3, 2024, pp. 272-285.

Ramsland, Katherine. “The Dark Side of AI.” Psychology Today, 23 May 2026, https://www.psychologytoday.com/us/blog/shadow-boxing/202605/the-dark-side-of-ai. 17 July 2026.

Sano-Franchini, Jennifer, Megan McIntyre, and Maggie Fernandes. “Refusing GenAI in Writing Studies: A Quickstart Guide. Refusing Generative AI in Writing Studies." Refusing Generative AI in Writing Studies. 2025.

“Scientific Consensus.” NASA. https://science.nasa.gov/climate-change/scientific-consensus/. 17 July 2026.

Vee, Annette. “Large Language Models Write Answers.” Composition Studies, vol. 51, no. 1, 2023, pp. 176-181.

Wei, Marlynn. “The Emerging Problem of ‘AI Psychosis.’” Psychology Today, 27 Nov. 2025. https://www.psychologytoday.com/us/blog/urban-survival/202507/the-emerging-problem-of-ai-psychosis. 17 July 2026.

Whittaker, Zack. “For Privacy and Security, Think Twice Before Granting AI Access to your Personal Data.” TechCrunch, 19 July 2025. https://techcrunch.com/2025/07/19/for-privacy-and-security-think-twice-before-granting-ai-access-to-your-personal-data/. 17 July 2026.

Yañez-Barnuevo, Miguel. “Data Center Buildout is Hungry for Fossil Fuels.” Environmental and Energy Study Institute, 26 Jan. 2026. https://www.eesi.org/articles/view/data-center-buildout-is-hungry-for-fossil-fuels. 17 July 2026.

---. “Data Centers and Water Consumption.” Environmental and Energy Study Institute, 25 Jan. 2025. https://www.eesi.org/articles/view/data-centers-and-water-consumption. 17 July 2026.