Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LLM or Human? Perceptions of Trust and Information Quality in Research Summaries

Nil-Jana Akpinar, Sandeep Avula, CJ Lee, Brandon Dang, Kaza Razat, Vanessa Murdock

TL;DR

This study investigates how ML-expert readers perceive trust and information quality in abstracts produced or edited by Large Language Models. Using a mixed-design online experiment with three abstract types (human-written, LLM-generated, LLM-edited) drawn from arXiv ML papers, the authors examine detection ability, effects of actual and perceived LLM involvement, and reader orientations toward AI-assisted writing. Key findings show readers cannot reliably identify LLM involvement, yet belief in LLM use significantly shifts judgments, with LLM-edited abstracts often rated most favorable when disclosure is present. The work identifies three reader orientations—Disclosure Advocates, Pragmatic Skeptics, and Optimists—and discusses implications for disclosure policies and the design of AI-assisted scientific writing. Overall, transparency boosts trust in AI-assisted abstracts among ML experts, pointing to nuanced norms needed for credible scientific communication.

Abstract

Large Language Models (LLMs) are increasingly used to generate and edit scientific abstracts, yet their integration into academic writing raises questions about trust, quality, and disclosure. Despite growing adoption, little is known about how readers perceive LLM-generated summaries and how these perceptions influence evaluations of scientific work. This paper presents a mixed-methods survey experiment investigating whether readers with ML expertise can distinguish between human- and LLM-generated abstracts, how actual and perceived LLM involvement affects judgments of quality and trustworthiness, and what orientations readers adopt toward AI-assisted writing. Our findings show that participants struggle to reliably identify LLM-generated content, yet their beliefs about LLM involvement significantly shape their evaluations. Notably, abstracts edited by LLMs are rated more favorably than those written solely by humans or LLMs. We also identify three distinct reader orientations toward LLM-assisted writing, offering insights into evolving norms and informing policy around disclosure and acceptable use in scientific communication.

LLM or Human? Perceptions of Trust and Information Quality in Research Summaries

TL;DR

This study investigates how ML-expert readers perceive trust and information quality in abstracts produced or edited by Large Language Models. Using a mixed-design online experiment with three abstract types (human-written, LLM-generated, LLM-edited) drawn from arXiv ML papers, the authors examine detection ability, effects of actual and perceived LLM involvement, and reader orientations toward AI-assisted writing. Key findings show readers cannot reliably identify LLM involvement, yet belief in LLM use significantly shifts judgments, with LLM-edited abstracts often rated most favorable when disclosure is present. The work identifies three reader orientations—Disclosure Advocates, Pragmatic Skeptics, and Optimists—and discusses implications for disclosure policies and the design of AI-assisted scientific writing. Overall, transparency boosts trust in AI-assisted abstracts among ML experts, pointing to nuanced norms needed for credible scientific communication.

Abstract

Large Language Models (LLMs) are increasingly used to generate and edit scientific abstracts, yet their integration into academic writing raises questions about trust, quality, and disclosure. Despite growing adoption, little is known about how readers perceive LLM-generated summaries and how these perceptions influence evaluations of scientific work. This paper presents a mixed-methods survey experiment investigating whether readers with ML expertise can distinguish between human- and LLM-generated abstracts, how actual and perceived LLM involvement affects judgments of quality and trustworthiness, and what orientations readers adopt toward AI-assisted writing. Our findings show that participants struggle to reliably identify LLM-generated content, yet their beliefs about LLM involvement significantly shape their evaluations. Notably, abstracts edited by LLMs are rated more favorably than those written solely by humans or LLMs. We also identify three distinct reader orientations toward LLM-assisted writing, offering insights into evolving norms and informing policy around disclosure and acceptable use in scientific communication.
Paper Structure (42 sections, 3 figures, 5 tables)

This paper contains 42 sections, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Trust, Information Quality, and Measurement
  4. Trust in LLM-Generated Text and Summaries
  5. Human and Algorithmic Detection of LLM Involvement
  6. Effects of Disclosing AI Involvement
  7. Methods
  8. Study Design
  9. Participants
  10. Materials
  11. Paper Selection and Abstract Data
  12. LLMs, Prompts, and Abstract Types
  13. Measures and Survey Procedure
  14. Procedure
  15. LLM authorship attribution
  16. ...and 27 more sections

Figures (3)

  • Figure 1: Probability of selecting each abstract type by information condition. Error bars indicate 95% confidence intervals. When LLM involvement is revealed, participants show a strong preference towards LLM-edited abstracts.
  • Figure 2: Participants in PCA space (PC1 vs. PC2), colored by PC3 (sequential blue scale). Participants fall into three clusters based on their orientation towards LLM-generated content with straight-sided convex-hull enclosures per cluster.
  • Figure 3: Linguistic and structural characteristics of abstracts across different generation methods.