Human-made vs. AI-generated: how provenance labels drive strategic curation via perceived effort.
Level 2 - randomized trial
Randomized between-subjects factorial experiment with human participants
PubMed 42317700 · doi:10.3389/fpsyg.2026.1840483
What was done
A 3 (label type: human-made vs. AI-generated vs. unlabeled) × 2 (content type: eudaimonic vs. hedonic) between-subjects factorial experiment was conducted with 618 short-form video platform users. The authors evaluated how provenance labels affect perceived creator effort, beliefs about algorithmic intervention, and strategic curation intentions through rational and normative pathways.
What was found
The abstract reports directional findings without quantitative values, effect sizes, or test statistics: - AI-generated labels significantly reduced perceived creator effort, whereas human-made labels did not differ from unlabeled content (indicating a default human-made assumption). - Perceived creator effort increased strategic curation intentions via both rational and normative pathways, but AI-generated labels weakened both pathways. - Effort devaluation from AI labeling was context-independent across content types (eudaimonic vs. hedonic). - Exploratory analysis showed that higher algorithmic knowledge was associated with lower intervention intentions.
Why it matters
This study shows that AI provenance labels function as value cues that devalue perceived creator effort and reduce user motivation to curate algorithmic feeds, rather than acting solely as risk warnings.
Limits
The abstract provides no numerical data, confidence intervals, or exact p-values. Outcomes were measured as self-reported behavioral intentions and subjective beliefs rather than actual platform behavior. Participant demographics and sampling method are not described in the abstract.