CoChat

Large language models and the problem of rhetorical debt

Open access
MacArthur, Marit · 2025 · AI & SOCIETY

The DOI matches its Crossref record; 50 of the 50 references it cites were found in a registry or an index; 3 of the 56 sources its claims cite were checked for support.

Abstract

This article offers broadly useful guidance for society’s adaptation to the omnipresence of generative AI, with implications for every profession and academic discipline that involves writing or coding (recognized by some as a form of writing). Offering an interdisciplinary perspective grounded in the digital humanities, software development and writing across the curriculum, and building on performance historian Christopher Grobe’s research on the role of arts and humanities expertise in AI development, I offer redefinitions of training data and prompt engineering . These essential yet misleading terms obscure the critical roles that humanities-based expertise has played in the development of GPTs and must play in guiding society’s adaptation to generative AI. I also briefly review scholarship on what constitutes “writing” and what it means to teach writing. Next, I reflect on long-terms trends, in professional software development, of code sharing and reliance on automation, and the likely impact of imposing similar practices in professional writing. After identifying the fundamental problem of rhetorical debt and outlining its consequences, I further motivate my argument, in relation to the new economic value of expert writing. This new economic value necessitates a revaluation of the humanities—not only by computer science, the tech industry, and schools and universities, but by humanists themselves.

Overview

CoChat Summary

TL;DR

This conceptual article argues that generative AI’s “training data” is human expertise captured in writing and that “prompt engineering” is really prompt writing, requiring rhetorical judgment. It introduces “rhetorical debt” as the writing analogue of technical debt and calls for revaluing humanities-based writing education as AI increases the economic value of expert writing.

Key findings

  • The paper redefines “training data” as human expertise captured in writing that must be continually developed and curated by humans, and “prompt engineering” as prompt writing that details the immediate rhetorical situation—purpose, audience, genre, and local context—which an LLM cannot access without human guidance.
  • It identifies “rhetorical debt” as accruing when writers in haste fail to critically edit text to suit audience, purpose, genre, and context, lack the expertise to do so, or uncritically use AI-generated text; this parallels software technical debt and is worsened by automation bias and a “fluency fallacy.”
  • The article cites evidence of AI’s dependence on expert human writing: AI companies hired college graduates at 20–40 USD/h to write for and train AI; Taylor & Francis allowed Microsoft to use authors’ work as training data for 10 million USD in the first year; and the 2025 NACE survey ranked written communication skills (77.1%) among the top in-demand skills, behind problem-solving (88.3%) and teamwork (81%).
  • It warns that fluent but inappropriate AI output can escape expert attention, citing Stack Overflow’s December 2022 ban on AI-generated text and later human-verified Answer Assistant, Boeing 737 Max 9 disasters that killed 346 people and cost at least $20 billion, Tesla recalls of 2 million vehicles in December 2023 and 2.2 million in February 2024, and hallucination rates as high as 79% in some reasoning systems trained via reinforcement learning.

Methods

  • This is a conceptual, interdisciplinary argument grounded in the digital humanities, software development, and writing across the curriculum, building on Christopher Grobe’s research on arts and humanities expertise in AI development.
  • The paper reviews scholarship on writing and writing instruction, reflects on long-term trends in professional software development, and develops the concept of rhetorical debt through illustrative examples and secondary sources.
  • No empirical sample, dataset, or formal analysis is reported; the author draws on cross-disciplinary experience and cited literature rather than primary data collection.

Limitations

  • The paper states that it restricts its consideration of training data to text, excluding images, video, and curated numerical datasets, and that creative writing is outside its scope; it also says reviewing centuries of writing scholarship is beyond its scope.
  • No explicit limitations section is present. Evident from the text: the argument is not empirically tested, relies on illustrative cases and secondary literature, and makes predictive claims about future workplace and educational effects that are interpretive rather than experimentally validated.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1007/s00146-025-02403-w

Cites

50 of 50 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

3 Valid · 4 Not valid · 44 Not checked · 56 Own results and interpretations

20 more claims

Create a free account to see all 107 claims and to verify the 53 references still open.

See all claims — free account
  • It may state two separate claims as one. 39
  • Probably not a substantial claim. 61
  • The quote only partly states this claim. 2

Create a free account to see these claims.

Citations

Cites

Show all 50 references

Research works printed but not linked: 10

  1. Battersby M (2024) Academic authors “shocked” after Taylor & Francis sells access to their research to Microsoft AI. The Bookseller. https://www.thebookseller.com/news/academic-authors-shocked-after-taylor--francis-sells-access-to-their-research-to-microsoft-ai. Accessed 4 Aug 2024
  2. Bitzer LF (1968) The rhetorical situation. Philos Rhet 1(1):1–14. http://www.jstor.org/stable/40236733
  3. Ecclesiastes 1:9. The New Oxford Annotated Bible (3rd ed.) (450~150 BCE, 2001 CE). Coogan M et al (eds) Oxford University Press
  4. Glynn A (2024) Suspected undeclared use of artificial intelligence in the academic literature: an analysis of the Academ-AI Dataset. arXiv.org. https://arxiv.org/abs/2411.15218 · arXiv
  5. Gray K, Koncz A (2025) The attributes employers look for on new grad resumes—and how to showcase them. National Association of Colleges and Employers. https://www.naceweb.org/about-us/press/the-attributes-employers-look-for-on-new-grad-resumes-and-how-to-showcase-them. Accessed 20 Mar 2025
  6. Murphy JJ (ed) (1972) A synoptic history of classical rhetoric. Random House, New York
  7. Oxford Word of the Year: Our 2024 Shortlist (2024) Oxford University Press. https://corp.oup.com/word-of-the-year/#shortlist-2024. Accessed 28 Jan 2025
  8. Perelman C, Oblrechts-Tyteca L (1958, 1969) The new rhetoric: a treatise on argumentation, trans. Wilkinson J and Weaver P. University of Notre Dame Press
  9. Rohman DG (1965) Pre-writing: the stage of discovery in the writing process. Coll Compos Commun 16(2):106–112
  10. Tenen DY (2024) Literary theory for robots: how computers learned to write. W.W. Norton, New York

Cited by

See all 8

Metadata

Authors
MacArthur, M.
Affiliations
University of California, Davis (US)
Corresponding author
MacArthur, Marit
Venue
AI & SOCIETY
Published
2025-06-25
Volume
40
Issue
8
Pages
6425-6438
ISSN
  • 0951-5666 (ISSN-L)
  • 1435-5655
OpenAlex
W4411661084
Cited by
8
References
44
Citations by year
2026: 5 · 2025: 3
FWCI
8.28
Type
journal article
Language
en
Open access
Licence
Locations
  • AI & Society (publishedVersion, cc-by)
  • SSRN Electronic Journal (acceptedVersion)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
  • Social Sciences › Arts and Humanities › Literature and Literary Theory › Digital Humanities and Scholarship
  • Social Sciences › Arts and Humanities › History and Philosophy of Science › Cybernetics and Technology in Society
Keywords
generative AI, large language models, training data, prompt engineering, writing instruction, professional software development, code sharing, automation

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.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-09-30, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-09-30 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

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

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Large language models and the problem of rhetorical debt. https://app.cochat.ai/papers/10.1007/s00146-025-02403-w
The paper
MacArthur, M. (2025). Large language models and the problem of rhetorical debt. AI & SOCIETY. https://doi.org/10.1007/s00146-025-02403-w

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.