EUVIMEDEUVIMED entdecken
MetaanalyseOpen Access · CC BY 4.0

Does an AI authorship label change literary evaluation? A meta-analysis of randomized label experiments in poetry and fiction

Elfrine Hingis Manampiring

Journal of Language and Literature Inquiry · 15.11.2026 · 1 (2) · 70-91

DOI: 10.55942/jlli.v1i2.2365 ↗

Kurz gefasst

Metaanalyse in der Zeitschrift Journal of Language and Literature Inquiry, online erschienen am 15.11.2026. Verfasst von Elfrine Hingis Manampiring. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Readers increasingly meet poems and stories labeled as written by artificial intelligence (AI), yet it remains unclear how much such a label, independently of the text, changes literary evaluation. This meta-analysis synthesized randomized experiments in which identical or equivalent poems and stories were attributed to AI or to a human author. Semantic Scholar searches, citation chasing, and searches for preprints identified nine studies with 12 independent samples. Hedges g was computed as the AI-label minus human-label difference, pooled with random-effects models using restricted maximum likelihood and Knapp-Hartung confidence intervals, and appraised for risk of bias and certainty. Across seven samples from six research teams (N = 3,870), the AI label lowered aesthetic evaluation, g = -0.28, 95% CI [-0.38, -0.17], with modest heterogeneity and a prediction interval below zero. Narrative transportation and absorption showed a similar reduction across six samples from four teams, g = -0.29 [-0.41, -0.18]. Effects on rated creativity were imprecise, and comparisons of AI labels with no label were inconsistent. The penalty on evaluation was robust to team-level aggregation, the exclusion of reports that were not peer reviewed, and leave-one-team-out analyses, although it was smaller for fiction than in the single poetry sample, and descriptive evidence from two teams suggested that science fiction weakened the penalty on engagement. Certainty was moderate for evaluation and low for engagement, mainly because the samples read short English or German texts in online adult samples. Disclosure therefore carries a measurable reception cost that belongs to the reading situation rather than to the text.

Zitieren

APA

Manampiring, E. H. (2026). Does an AI authorship label change literary evaluation? A meta-analysis of randomized label experiments in poetry and fiction. Journal of Language and Literature Inquiry, 1(2), 70-91. https://doi.org/10.55942/jlli.v1i2.2365

Vancouver

Manampiring EH. Does an AI authorship label change literary evaluation? A meta-analysis of randomized label experiments in poetry and fiction. Journal of Language and Literature Inquiry. 2026;1(2):70-91. doi:10.55942/jlli.v1i2.2365

BibTeX
@article{Manampiring2026,
  author = {Manampiring, Elfrine Hingis},
  title = {Does an AI authorship label change literary evaluation? A meta-analysis of randomized label experiments in poetry and fiction},
  journal = {Journal of Language and Literature Inquiry},
  year = {2026},
  volume = {1},
  number = {2},
  pages = {70--91},
  doi = {10.55942/jlli.v1i2.2365}
}
Kostenlos starten →