How do LLMs affect writing? Talks by visitors to the AI STORIES project

Programme

(Full abstracts and bios at bottom of page)

09:30-11:30 | Research talks (45 min + Q&A)

• Templated Genre Knowledge: The Orbital Arguments of LLMs Help Us Understand Genre. John Gallagher (external link) (Professor, Department of English, University of Illinois Urbana-Champaign)
  • John Gallagher's new book is AI Through the Experts’ Eyes: Communicating Complex Ideas (external link) (2026), which draws on more than 100 interviews with AI scientists and practitioners. He regularly writes insightful essays about LLM writing (external link).
• The Performative Archive: Large Language Models as Material Culture (Pedro Jacobetty (external link), Research Associate, Department of Educational Science, University of Potsdam)

11:30-12:30 | Lunch

Visitors are warmly welcome to bring their own lunch and join us for an informal shared break.

12:30-14:30 | Research talks (45 min + Q&A)

• How LLMs Distort Our Written Language (Marwa Abdulhai (external link), Postdoctoral Fellow, Princeton University, and Isadora White (external link), PhD Candidate, UC San Diego)
  • Marwa and Isadora will present their recent precent, which is a quantitative analysis of LLM-edited texts finding that the LLM revisions caused large changes in the content and meaning of the texts, even when the LLM was only prompted to edit grammar. The preprint is available at Arxiv, and you can read media reports about the research in Psychology Today (external link) and NBC News (external link).
  •  
• Potential Impact of LLM-Based Chatbots on Textual Features of Electronic Word-of-Mouth (Rashid Mustafin (external link) (PhD Candidate, Department of Professional and Intercultural Communication, Norwegian School of Economics (NHH))

Abstracts and Speaker Bios

John Gallagher: Templated Genre Knowledge: The Orbital Arguments of LLMs Help Us Understand Genre.

Abstract: Large language models do not make very good arguments. Instead, they orbit them, circling around meaning through various rhetorical forms, such as cataloguing, hypotaxis, and chains of three. This talk uses this orbital quality to argue that formation of genre is different for human than LLMs. Humans learn genres from the bottom up, that is, we accrue conventions one example at a time until patterns emerge from situated experience. LLMs invert this process. Trained on templated content, they hold the conventions of all genres simultaneously, in a kind of superposition, producing what I call templated genre knowledge: the abstraction of a genre with none of the interesting nuance of a singular instance. The result is writing and argumentation that feels correct on a skim yet dissolves under close reading. I thus argue that LLM-derived sentences require interpretation rather providing any meaning. The orbital arguments of LLMs, I conclude, provide strong evidence that rhetorical genres are lived recurrent social actions.

BIO: John R. Gallagher is professor of English at the University of Illinois, Urbana-Champaign, where he also is a faculty affiliate at the School of Information Sciences. He is the author of AI Through the Experts’ Eyes: Communicating Complex Ideas (2026), Case Study Research in the Digital Age (2024), and Update Culture and the Afterlife of Digital Writing (2020). He fuses qualitative inquiry with natural language processing and machine learning to study social media.

Pedro Jacobetty: The Performative Archive: Large Language Models as Material Culture

Abstract: In this talk, LLMs are approached as complex embedded systems rather than discrete technical objects via the concept of the performative archive: a system that recombines, predicts, and generates discourse rather than merely storing it. The analysis combines perspectives from complex-systems theory, social theory, semiotics, and critical theory. The argument moves through technical, semiotic, and political-economic registers as co-constituting each other. LLMs are allopoietic semiotic machines, whose effects are relationally and indexically produced without central symbolic control, while simultaneously functioning as engines of material and political centralization. The theoretical claims are grounded in technical interrogation and model interpretability research.  

BIO: Pedro Jacobetty is a research fellow at the Department of Educational Science, University of Potsdam. 

Rashid Mustafin: Potential Impact of LLM-Based Chatbots on Textual Features of Electronic Word-of-Mouth

Abstract: Electronic word-of-mouth (eWOM) is often regarded as a trustworthy source of information because it is meant to be produced by consumers who have experienced the product or service of interest. However, for many years digital platforms for consumer feedback have already been polluted by fake reviews. The launch of LLM-based chatbots made the production of fluent texts substantially easier, which is likely to exacerbate the problem of fake reviews. Estimating the degree and the nature of impact that the use of LLMs has had on consumer feedback can be approached by classifying each review individually with special detector models or by comparing larger corpora that were collected before and after the introduction of LLMs to the wider public. More attention is paid to the former method, while the latter has been used primarily for studying the changes in academic discourse. This study applies corpus analysis to consumer reviews in order to identify examples of potential LLM influence that are harder to spot when examining individual texts but are observable on a larger scale. The preliminary results show that the reviews published after 2022 exhibit more features of LLM-generated texts at the lexical, syntactic and discursive levels. These findings have implications for research practices involving eWOM data and for discussions on trust in digital information environments. 

BIO: Rashid Mustafin is a PhD student in Digital Text Analysis for Market Research at the Norwegian School of Economics. His research interests include the application of language technologies in business, linguistically informed natural language processing and digital methods in humanities and social sciences. He has experience in sentiment analysis, stylometry, computational analysis of social media communication, cognitive and usage-based linguistics. 

Marwa Abdul Hai and Isadora White: How LLMs Distort Our Written Language

Abstract: Large language models (LLMs) are used by over a billion people globally, most often to assist with writing. In this work, we demonstrate that LLMs not only alter the voice and tone of human writing, but also consistently alter the intended meaning. First, we conduct a human user study to understand how people actually interact with LLMs when using them for writing. Our findings reveal that extensive LLM use led to a nearly 70% increase in essays that remained neutral in answering the topic question. Significantly more heavy LLM users reported that the writing was less creative and not in their voice. Next, using a dataset of human-written essays that was collected in 2021 before the widespread release of LLMs, we study how asking an LLM to revise the essay based on the human-written feedback in the dataset induces large changes in the resulting content and meaning. We find that even when LLMs are prompted with expert feedback and asked to only make grammar edits, they still change the text in a way that significantly alters its semantic meaning. We then examine LLM-generated text in the wild, specifically focusing on the 21% of AI-generated scientific peer reviews at a recent top AI conference. We find that LLM-generated reviews place significantly less weight on clarity and significance of the research, and assign scores that, on average, are a full point higher. These findings highlight a misalignment between the perceived benefit of AI use and an implicit, consistent effect on the semantics of human writing, motivating future work on how widespread AI writing will affect our cultural and scientific institutions.

BIO: Isadora White is a PhD Student at the University of California, San Diego, and a researcher at Microsoft Research. Isadora's research spans training techniques to improve LLM capabilities to more recent work on AI security, ethics, and societal impacts. Recently, Isadora and Marwa's preprint "How LLMs Distort Our Written Language (external link)" was featured in NBC News and Psychology Today (external link), and she has co-authored papers in venues such as EMNLP, ICLR, and ICML.

BIO: Marwa Abdulhai is an incoming Presidential Postdoctoral Fellow at Princeton University. She completed her PhD in Computer Science at UC Berkeley advised by Professor Sergey Levine with Professor Natasha Jaques as a close collaborator. Her research focuses enabling AI agents to better understand people and their interactions to build both safe and more AI capable systems. This includes improving the performance of existing large language models (LLMs) for multi-turn dialogue interactions, understanding how to protect against deception in AI systems, and exploring how AI can serve as a useful tool for social science research. Her research has been supported by the Quad Fellowship, AI Policy Hub, Open AI Research, and Cooperative AI PhD Fellowship.