Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 65 of the 67 references it cites were found in a registry or an index, and 2 are not checked yet; none of the 52 sources its claims cite is checked for support yet.

Epistemic Trust in Generative AI as an Information Source: Development and Validation of the Trust in AI-Epistemic Scale (TAI-E)

Open access
Balázs, László · 2026 · Social Sciences

Abstract

As generative AI increasingly mediates everyday information, users must judge not only whether a claim is true but whether the system conveying it is a legitimate source of belief. This article develops and validates the Trust in AI-Epistemic Scale (TAI-E), measuring epistemic trust in generative AI across three dimensions: attribution of epistemic authority, perceived epistemic reliability, and suspension of critical judgment. In a survey of part-time higher-education students (N = 412; ages 18–63), the three-factor structure was replicated across a calibration–validation split (CFI = 0.944; RMSEA = 0.085; subscale α = 0.84–0.87). Granting a system epistemic authority and suspending critical judgment toward it behaved as partly opposing stances, so the scale captures a configuration rather than a single trust score. Self-reported reflective processing accompanied higher authority and reliability attributions together with greater critical engagement. Most strikingly, self-reported reflectiveness barely tracked behavioural cognitive reflection (r = 0.14), and only the self-report predicted trust—suggesting it reflects epistemic self-image as much as reflective capacity. Relations among constructs are reported as construct-validity associations rather than causal effects. The TAI-E provides a psychometrically grounded tool for studying epistemic trust, AI literacy, and well-founded reliance on AI.

Checked 2026-11-18 · 2 references not checked yet

Overview

CoChat Summary

TL;DR

This paper develops and validates the Trust in AI-Epistemic Scale (TAI-E), a 12-item instrument measuring epistemic trust in generative AI across three dimensions: attribution of epistemic authority, perceived epistemic reliability, and suspension of critical judgment. In a sample of 412 part-time higher-education students, the three-factor structure replicated across a calibration–validation split, but the dimensions did not cohere into a single composite, and self-reported reflectiveness—not behavioural cognitive reflection—carried the association with trust.

Key findings

  • The three-factor structure replicated across a calibration (EFA, n = 206) and validation (CFA, n = 206) split: EFA explained 66.7% of variance with loadings λ = 0.59–0.92; CFA fit was acceptable-to-marginal in the validation half (CFI = 0.916; TLI = 0.891; RMSEA = 0.099) and improved in the full sample (CFI = 0.944; TLI = 0.927; RMSEA = 0.085; SRMR = 0.048).
  • Subscale reliabilities were good (α = 0.84–0.87; ω = 0.89–0.91), but no overall composite was supported: Suspension of Critical Judgment correlated negatively with Authority (r = −0.63) and Reliability (r = −0.48), and the 12-item total score had poor internal consistency (α = 0.47, ω = 0.56).
  • Self-reported reflective information processing correlated positively with Authority (r = 0.35) and Reliability (r = 0.32) and negatively with Suspension (r = −0.33), all p < 0.001—a configural rather than monotonic pattern.
  • Self-reported reflectiveness correlated only weakly with behavioural CRT-3 performance (r = 0.14, p = 0.005, ~2% shared variance); CRT-3 was at most weakly related to Authority (r = 0.11) and unrelated to Reliability (r = 0.05) and Suspension (r = 0.00).
  • Criterion relations: more frequent AI use predicted lower Authority (β = −0.23, p < 0.001) and lower Reliability (β = −0.15, p = 0.002) but not Suspension (β = 0.07, p = 0.141); age predicted greater critical engagement (Suspension β = 0.15, p = 0.002) but was unrelated to Authority and Reliability. Sense of coherence was essentially unrelated to the trust dimensions (|r| ≤ 0.15).

Methods

  • Instrument development: 30 initial items reduced to 22 after review by three experts (psychology, computer science, linguistics; 8 items discarded, 73.3% retention), then to 12 items via EFA using principal axis factoring with oblimin rotation.
  • Sample: online questionnaire (Google Forms) distributed to correspondence/part-time students at several Hungarian higher-education institutions in the first three weeks of April 2026; inclusion required age ≥ 18, Hungarian proficiency, and at least occasional generative AI use. Final analytic sample N = 412 (20 cases, 4.6%, excluded), ages 18–63 (M = 36.56), 62% women, with overrepresentation in engineering (24.8%) and IT (18.9%).
  • Measures and analysis: 6-point Likert TAI-E items (Suspension reverse-coded); SOC-13 (α = 0.78); a 5-item self-report reflective information processing scale (α = 0.75); single-item AI-use frequency; age; and the three-item CRT. Analyses included EFA/CFA on a random split, reliability, AVE/Fornell-Larcker and HTMT discriminant validity, Pearson correlations, and linear regressions with HC3 robust standard errors. Construct-validity analyses were exploratory and not preregistered; the CRT comparison was added post hoc during revision.

Limitations

  • Stated: cross-sectional design permits only associational, not causal, interpretation; convenience sampling of distance-learning students with high AI exposure (95%) and engineering/IT overrepresentation (43.7%) limits generalizability; all variables were self-reported in one questionnaire, raising common method bias concerns; the calibration–validation split came from a single dataset rather than independently collected samples, and the CFA RMSEA (0.085, validation upper 90% CI approaching 0.10) justifies refinement and replication; the expert panel lacked a philosopher specializing in epistemology, philosophy of technology, or AI ethics.
  • Evident from the text: the Suspension of Critical Judgment subscale captures a reported disposition rather than observed suspension, and distinguishing genuine suspension from ordinary belief revision would require a behavioural paradigm; the self-report reflective processing scale retained an item with a low loading (λ = 0.323, communality = 0.104); the Reliability subscale's AVE (0.49) fell marginally below the 0.50 benchmark; and the CRT-3 reliability was modest (α = 0.60).

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

Integrity signals

Registry

Crossref · 10.3390/socsci15070433

Cites

65 of 67 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Three-factor confirmatory factor model of the TAI-E scale (standardized estimates).
Figure 1 Three-factor confirmatory factor model of the TAI-E scale (standardized estimates).CC-BY · From the PDF

Verification

Claims the paper makes

0 Valid · 0 Not valid · 23 Not checked · 112 Own results and interpretations

44 more claims

Create a free account to see all 135 claims and to verify the 43 references still open.

See all claims — free account
  • Probably not a substantial claim. 29
  • It may state two separate claims as one. 68
  • The quote only partly states this claim. 2

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Show all 64 references

Research works printed but not linked: 2

  1. Couldry, Nick, and Hepp, Andreas (2017). The Mediated Construction of Reality, Polity Press.
  2. Gerlich, Michael (, January May). The Societal Perceptions of Artificial Intelligence and Its Impact on Marketing in 2024. Paper presented at 5th World Conference on Business, Management, Finance, Economics and Marketing, Vienna, Austria.

Cited by

See all 2

Metadata

Authors
Balázs, L.
Affiliations
University of Dunaújváros (HU)
Corresponding author
Balázs, László
Venue
Social Sciences
Publisher
MDPI AG
Published
2026-07-01
Volume
15
Issue
7
Pages
433
ISSN
2076-0760 (ISSN-L)
OpenAlex
W7166889414
Cited by
2
References
64
Citations by year
2026: 2
FWCI
8.32
Type
journal article
Language
en
Open access
Locations
  • Social Sciences (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
Funders
University of Dunaújváros
Topics
  • Social Sciences › Psychology › Developmental and Educational Psychology › Educational Strategies and Epistemologies
  • Social Sciences › Arts and Humanities › Philosophy › Epistemology, Ethics, and Metaphysics
  • Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
Keywords
epistemic trust, generative AI, epistemic authority, epistemic reliability, reflective processing, cognitive reflection, AI literacy, scale development, university students

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.

A language model drew the claims from the paper on 2026-11-18, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-11-18 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

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

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.