pith. sign in

arxiv: 2605.02649 · v1 · submitted 2026-05-04 · 💻 cs.HC

Dramaturgies of Deception: AI Humanizers and the Performance of Legitimacy in Higher Education Assessment

Jasper Roe (1) , Mike Perkins (2) , Peter Bannister (3) , Leon Furze (4) , James Wood (1) ((1) Durham University , United Kingdom , (2) British University Vietnam , Vietnam
show 4 more authors
(3) International University of La Rioja Spain (4) Deakin University Australia)
This is my paper

Pith reviewed 2026-05-08 19:10 UTC · model grok-4.3

classification 💻 cs.HC
keywords AI humanizershigher education assessmentacademic integrityperformative assessmentGoffman dramaturgyAI detectionstructural reform
0
0 comments X

The pith

AI humanizers reveal a self-reinforcing cycle of surveillance and deception in university assessments that calls for structural reform rather than improved detectors.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper catalogs 55 AI humanizer websites and performs a detailed multimodal critical discourse analysis on three of them, applying Goffman's dramaturgical lens to show how these services market themselves. They delete references to misconduct, frame their use as a rational response to flawed institutional surveillance and AI detectors, and create an aura of advanced technology with implied endorsements from universities and companies. This positions the services as a legible symptom rather than an isolated problem. A reader would care because the findings indicate that escalating technological arms races between detectors and humanizers will not resolve underlying issues of assessment design and trust.

Core claim

Through systematic search and custom rubric applied to 55 sites plus in-depth analysis of three, the paper establishes that humanizers perform similar functions: discursive absence of misconduct, rational framing as a defensible response to surveillance, and mystification via technology and implied institutional endorsement. Humanizer services therefore function as a diagnostic signal, a node in a feedback loop of performative assessment where policies demanding independent authorship meet tools that simulate it and detectors that attempt to catch the simulation.

What carries the argument

Goffman's dramaturgical account of performance and legitimacy applied to the marketing language, design, and claims of AI humanizer sites, which shows how they stage an identity of helpful, legitimate assistance.

If this is right

  • Humanizer services are widely available in both free and premium forms and perform consistent functions across sites.
  • They delete or avoid discussion of academic misconduct while presenting AI humanization as a legitimate counter to institutional overreach.
  • The cycle of detection and circumvention will persist unless assessment practices themselves change.
  • Technological solutionism focused on better detectors will likely accelerate demand for humanizers rather than reduce it.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the cycle holds, universities investing heavily in AI detectors may inadvertently drive more students toward humanizers and similar tools.
  • The findings suggest a need to examine how assessment criteria that emphasize process over product could reduce the incentive for simulation.
  • Similar dramaturgical analysis could be applied to other emerging services that help users simulate compliance in regulated environments.

Load-bearing premise

The observed patterns from the purposive sample of three sites and the broader catalog of 55 accurately represent how all such services operate and appeal to users without substantial selection bias or interpretive effects.

What would settle it

A study that surveys a large number of students using humanizers or analyzes actual submission patterns and finds that most users do not cite evasion of detection or view the services as a rational response to flawed systems would falsify the diagnostic-signal interpretation.

read the original abstract

Artificial intelligence (AI) has disrupted assessment in higher education and accelerated a cycle of compounding performances. Institutional policies demand the demonstration of independent authorship, while commercial AI-enabled services allow students to simulate independent thought and writing. This has led to enhanced institutional surveillance, including AI detectors, which are subsequently circumvented using other technologies. AI humanizers, internet-based services that alter AI-generated text to avoid automated or human detection, are a recent symptom of this performative cycle. Little is known about how these services operate, how they appeal to users, and what they imply for educational assessment and integrity. This paper presents an exploratory, systematic investigation of AI humanizer websites, framed through Goffman's sociological account of dramaturgy. Using a systematic search and custom rubric, we cataloged 55 humanizer sites, assessed their performance of identity, and conducted an in-depth multimodal critical discourse analysis of a purposive sample of three sites. Findings show that humanizers are readily available, offer free and premium paid services, and appear to perform similar functions. These include the deletion and discursive absence of misconduct, the framing of AI humanization as a rational and defensible response to surveillance and flawed detection, and appeals to mystification through advanced technology and implied endorsement by universities and corporations. We argue that humanizer services should be viewed as a diagnostic signal: a legible node in a feedback loop of performative assessment. Disrupting this cycle requires structural assessment reform rather than technological solutionism.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper claims that AI humanizer services—commercial tools that modify AI-generated text to evade detection—form part of a performative feedback loop in higher education assessment. Drawing on Goffman's dramaturgical theory, the authors conduct a systematic search to catalog 55 humanizer websites, apply a custom rubric to assess their identity performance, and perform multimodal critical discourse analysis on a purposive sample of three sites. Key findings include the services' deletion of misconduct references, framing of use as a rational response to flawed AI detectors and institutional surveillance, and appeals to technological mystification plus implied university/corporate endorsement. The central argument is that these services act as a diagnostic signal of deeper assessment problems, requiring structural reform rather than technological fixes like improved detectors.

Significance. If the interpretive patterns hold, the work usefully extends sociological analysis of AI in education by treating humanizers as legible symptoms of performative assessment rather than isolated technical artifacts. The systematic catalog of 55 sites supplies an empirical foundation that future studies can build upon, and the application of Goffman's framework provides a coherent theoretical lens for analyzing marketing language and user appeals. This challenges technological solutionism and could inform policy discussions on assessment redesign.

major comments (2)
  1. [Methods] Methods section: The criteria used to select the purposive sample of three sites from the catalog of 55 are not specified. This is load-bearing for the central claim because the multimodal critical discourse analysis of these sites is used to identify recurring functions (deletion of misconduct, rational-response framing, mystification) that are then generalized to characterize humanizer services as a whole and to support the diagnostic-signal conclusion.
  2. [Findings and Discussion] Findings and Discussion sections: The paper states that the three sites 'appear to perform similar functions' and extends this to the broader catalog and beyond, yet provides no description of inter-rater reliability, rubric validation procedures, or steps taken to mitigate researcher interpretation bias in the discourse analysis. This weakens the move from specific observations to the general claim that humanizers reliably signal a performative assessment loop.
minor comments (2)
  1. [Abstract and Methods] Abstract and Methods: The custom rubric is mentioned but its development, items, and application to the full catalog of 55 sites are not detailed, which would aid reproducibility even in an exploratory study.
  2. [Discussion] The paper could clarify in the Discussion how the dramaturgical framing distinguishes between service marketing claims and actual user practices, as the analysis relies primarily on website content.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their careful reading and constructive feedback, which highlights important areas for strengthening the methodological transparency of our exploratory study. We address each major comment below and commit to revisions that enhance clarity without altering the core interpretive claims.

read point-by-point responses

  1. Referee: [Methods] Methods section: The criteria used to select the purposive sample of three sites from the catalog of 55 are not specified. This is load-bearing for the central claim because the multimodal critical discourse analysis of these sites is used to identify recurring functions (deletion of misconduct, rational-response framing, mystification) that are then generalized to characterize humanizer services as a whole and to support the diagnostic-signal conclusion.

    Authors: We agree that explicit selection criteria are necessary to support the extension from the sample to broader patterns. The manuscript describes the sample as purposive but does not detail the decision rules. In the revised version, we will add a subsection specifying that the three sites were selected from the catalog of 55 on the basis of (1) high visibility in initial search results for 'AI humanizer' and related terms, (2) representation of both free and premium service models, and (3) diversity in the range of marketing claims presented on their homepages. This addition will make the sampling logic transparent and directly address the concern about generalizability. revision: yes

  2. Referee: [Findings and Discussion] Findings and Discussion sections: The paper states that the three sites 'appear to perform similar functions' and extends this to the broader catalog and beyond, yet provides no description of inter-rater reliability, rubric validation procedures, or steps taken to mitigate researcher interpretation bias in the discourse analysis. This weakens the move from specific observations to the general claim that humanizers reliably signal a performative assessment loop.

    Authors: We acknowledge that the current manuscript lacks explicit discussion of analytical procedures for ensuring rigor in the multimodal critical discourse analysis. As an interpretive, team-based study, the analysis involved iterative joint review by all authors to develop and apply the custom rubric and to reach consensus on thematic patterns; however, these steps are not documented. In the revision, we will expand the Methods section to describe rubric development, the collaborative coding process, reflexive practices used to surface potential interpretive biases, and the limitations of generalizing from a small purposive sample. We will also qualify the language in Findings and Discussion to emphasize that the observed functions are illustrative rather than statistically representative, while retaining the argument that the patterns function as a diagnostic signal. revision: yes

Circularity Check

0 steps flagged

No circularity: external theory applied to new observational data

full rationale

The paper performs a systematic search to catalog 55 humanizer sites, then applies Goffman's external dramaturgy framework to a purposive sample of three sites via multimodal critical discourse analysis. No equations, fitted parameters, or self-citations appear in the derivation. The central claim—that humanizers form a legible node in a performative assessment feedback loop requiring structural reform—is an interpretive conclusion drawn directly from the fresh website data and external sociological lens, without reduction to prior inputs by construction or self-referential definition.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim depends on the applicability of Goffman's dramaturgical account to commercial AI services and on the assumption that the sampled websites are representative of the broader phenomenon.

axioms (1)
  • domain assumption Goffman's sociological account of dramaturgy provides a valid and illuminating lens for analyzing commercial AI humanizer websites.
    The paper explicitly frames the entire investigation through this account without additional justification in the abstract.

pith-pipeline@v0.9.0 · 5625 in / 1383 out tokens · 36980 ms · 2026-05-08T19:10:32.745351+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

10 extracted references · 10 canonical work pages

  1. [1]

    Advance HE & Higher Education Policy Institute. (2025). Student academic experience survey

    work page 2025

  2. [2]

    https://www.advance-he.ac.uk/knowledge-hub/student-academic-experience- survey-2025 Bannister, P

    Advance HE. https://www.advance-he.ac.uk/knowledge-hub/student-academic-experience- survey-2025 Bannister, P. (2026). ParadAIse L0st? Higher Education Research & Development, 45(3), 847-856. https://doi.org/10.1080/07294360.2025.2586653 Bannister, P., Alcalde Peñalver, E., & Santamaría Urbieta, A. (2024). International students and generative artificial i...

    work page doi:10.1080/07294360.2025.2586653 2025

  3. [3]

    https://doi.org/10.1007/s40979-017-0022-5 Eaton, S. E. (2025). Neurodiversity and academic integrity: Toward epistemic plurality in a postplagiarism era. Teaching in Higher Education. https://doi.org/10.1080/13562517.2025.2583456 Ekman, P., & Friesen, W. V . (1969). Nonverbal leakage and clues to deception. Psychiatry, 32(1), 88–

    work page doi:10.1007/s40979-017-0022-5 2025

  4. [4]

    (with NetLibrary, Inc)

    https://doi.org/10.1080/00332747.1969.11023575 Fairclough, N. (with NetLibrary, Inc). (2003). Analysing discourse: Textual analysis for social research. Routledge. Fawns, T., Bearman, M., Dawson, P., Nieminen, J. H., Ashford-Rowe, K., Willey, K., Jensen, L. X., Damşa, C., & Press, N. (2025). Authentic assessment: From panacea to criticality. Assessment & ...

    work page doi:10.1080/00332747.1969.11023575 1969

  5. [5]

    https://doi.org/10.3389/feduc.2018.00067 Ofcom. (2024). Online nation 2024 report. https://www.ofcom.org.uk/siteassets/resources/documents/research-and-data/online- research/online-nation/2024/online-nation-2024-report.pdf?v=386238 Perkins, M., Furze, L., Roe, J., & MacVaugh, J. (2024). The Artificial Intelligence Assessment Scale (AIAS): A framework for ...

    work page doi:10.3389/feduc.2018.00067 2018

  6. [6]

    https://doi.org/10.1007/s40979-018-0036-7 Roe, J., & Perkins, M. (2022). What are automated paraphrasing tools and how do we address them? A review of a growing threat to academic integrity. International Journal for Educational Integrity, 18(1), Article

    work page doi:10.1007/s40979-018-0036-7 2022

  7. [7]

    https://doi.org/10.1007/s40979-022-00109-w Roe, J., & Perkins, M. (2023). 'What they're not telling you about ChatGPT': Exploring the discourse of AI in UK news media headlines. Humanities and Social Sciences Communications, 10, Article

    work page doi:10.1007/s40979-022-00109-w 2023

  8. [8]

    M., & McCarthy, G

    https://doi.org/10.1057/s41599-023-02282-w Rogerson, A. M., & McCarthy, G. (2017). Using Internet based paraphrasing tools: Original work, patchwriting or facilitated plagiarism? International Journal for Educational Integrity, 13(1),

    work page doi:10.1057/s41599-023-02282-w 2017

  9. [9]

    S., & Whiting, B

    https://doi.org/10.1007/s40979-016-0013-y Rowland, S., Slade, C., Wong, K. S., & Whiting, B. (2018). 'Just turn to us': The persuasive features of contract cheating websites. Assessment & Evaluation in Higher Education, 43(4), 652–665. https://doi.org/10.1080/02602938.2017.1391948 Saoula, O., Shamim, A., Mohd Suki, N., Ahmad, M. J., Abid, M. F., Patwary, ...

    work page doi:10.1007/s40979-016-0013-y 2018

  10. [10]

    https://doi.org/10.1007/s40979-023-00146-z Whitty, M. T. (2008). Revealing the ‘real’ me, searching for the ‘actual’ you: Presentations of self on an internet dating site. Computers in Human Behavior, Including the Special Issue: Integration of Human Factors in Networked Computing, 24(4), 1707–1723. https://doi.org/10.1016/j.chb.2007.07.002 Wood, J., & Pi...

    work page doi:10.1007/s40979-023-00146-z 2008