Answers are grounded in this paper and cite their sources.
Previous research strongly identifies transparency and AI labels as normative responses to emerging questions about the use of generative AI in news production. However, it remains unclear whether such labels effectively increase perceived credibility among audiences in Belgium (Dutch-speaking part). In a preregistered eye-tracking experiment (https://osf.io/3y5sv/overview?view_only=34db4958e23f4e7f92db34eb81323704) (N = 260), a 3 (AI label type: generic AI label, specific AI label, and the absence of an AI label) × 2 (modality: text vs. text-with-image news article) between-subjects design was employed in combination with levels of average fixation duration (low, medium, high) as a moderator. The results indicate that AI labels are often overlooked, and fixation duration is not a significant factor. Nonetheless, textual labelling often prompts readers to assume that the content was fully automated, without human oversight. Heatmaps illustrate that AI labels attract less attention than headlines. These findings underscore persistent challenges in making the use of AI visible to audiences and raise broader questions about whether AI disclosure strategies meaningfully enhance credibility in journalism.
No takes yet. Share an insight, caveat, or question.
D’haeseleer et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: