Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Understanding Reader Perception Shifts upon Disclosure of AI Authorship

Hiroki Nakano, Jo Takezawa, Fabrice Matulic, Chi-Lan Yang, Koji Yatani

TL;DR

This study investigates how readers' perceptions of an author shift when AI authorship is disclosed across six writing acts. Using a repeated-measures online design with $N=261$ participants and $990$ responses, the authors manipulate AI involvement from 0% to 100% and measure changes in trust, caring, competence, and likability, revealing a domain-specific “transparency penalty” that is strongest in social, interpersonal writing. They further show that higher AI literacy mitigates negative effects and that preserving perceptible human effort can sustain perceived authenticity. The findings offer design implications for context-sensitive transparency, human-effort visibility, and AI-literacy development to enable socially acceptable, transparent AI-assisted writing. Practically, the work argues for rethinking AI writing as a co-creative, relational practice rather than a pure automation problem, with implications for tool design and user education across cultures.

Abstract

As AI writing support becomes ubiquitous, how disclosing its use affects reader perception remains a critical, underexplored question. We conducted a study with 261 participants to examine how revealing varying levels of AI involvement shifts author impressions across six distinct communicative acts. Our analysis of 990 responses shows that disclosure generally erodes perceptions of trustworthiness, caring, competence, and likability, with the sharpest declines in social and interpersonal writing. A thematic analysis of participants' feedback links these negative shifts to a perceived loss of human sincerity, diminished author effort, and the contextual inappropriateness of AI. Conversely, we find that higher AI literacy mitigates these negative perceptions, leading to greater tolerance or even appreciation for AI use. Our results highlight the nuanced social dynamics of AI-mediated authorship and inform design implications for creating transparent, context-sensitive writing systems that better preserve trust and authenticity.

Understanding Reader Perception Shifts upon Disclosure of AI Authorship

TL;DR

This study investigates how readers' perceptions of an author shift when AI authorship is disclosed across six writing acts. Using a repeated-measures online design with participants and responses, the authors manipulate AI involvement from 0% to 100% and measure changes in trust, caring, competence, and likability, revealing a domain-specific “transparency penalty” that is strongest in social, interpersonal writing. They further show that higher AI literacy mitigates negative effects and that preserving perceptible human effort can sustain perceived authenticity. The findings offer design implications for context-sensitive transparency, human-effort visibility, and AI-literacy development to enable socially acceptable, transparent AI-assisted writing. Practically, the work argues for rethinking AI writing as a co-creative, relational practice rather than a pure automation problem, with implications for tool design and user education across cultures.

Abstract

As AI writing support becomes ubiquitous, how disclosing its use affects reader perception remains a critical, underexplored question. We conducted a study with 261 participants to examine how revealing varying levels of AI involvement shifts author impressions across six distinct communicative acts. Our analysis of 990 responses shows that disclosure generally erodes perceptions of trustworthiness, caring, competence, and likability, with the sharpest declines in social and interpersonal writing. A thematic analysis of participants' feedback links these negative shifts to a perceived loss of human sincerity, diminished author effort, and the contextual inappropriateness of AI. Conversely, we find that higher AI literacy mitigates these negative perceptions, leading to greater tolerance or even appreciation for AI use. Our results highlight the nuanced social dynamics of AI-mediated authorship and inform design implications for creating transparent, context-sensitive writing systems that better preserve trust and authenticity.

Paper Structure

This paper contains 44 sections, 2 figures, 11 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. AI-assisted Writing for Diverse Communicative Purposes
  4. Reader Perceptions of AI Disclosure
  5. Our Approach
  6. Method
  7. Experiment Design
  8. Procedure
  9. Pre-disclosure Evaluation
  10. Simulated AI Authorship Disclosure and Post-disclosure Evaluation
  11. Post-experimental Questionnaire
  12. Participants
  13. Data Analysis
  14. Quantitative Analysis
  15. Qualitative Analysis
  16. ...and 29 more sections

Figures (2)

  • Figure 1: The user interfaces of the experiment: (a) pre-rating and (b) post-rating phases. The original interface was in Japanese and the text shown here is translated for presentation, but the layout and visual design are preserved.
  • Figure 2: Act-level visualization of the regression analysis for all outcome variables. For visibility, the score of the perception shift was divided by the number of items to uniformize the range across the metrics. AIRatio shows a negative slope across all metrics. Notable findings: (1) Interact has a significantly negative intercept for TrustworthinessShift, CompetenceShift, CaringShift, and Likability. (2) Convince and Imagine have significantly positive intercepts for CompetenceShift. (3) Explore shows a significantly positive effect for CaringShift, and Imagine shows a significantly positive conditional effect with AIRatio for CaringShift.