Trust, Attribution, and Adoption: Creator Perspectives on AI Disclosure and Provenance

Sammanfattning

Generative artificial intelligence is increasingly used in visual content creation, including ideation, reference work, image generation, editing, and refinement. As AI becomes embedded in creative workflows, disclosure and provenance mechanisms are being introduced to communicate whetherand how AI was involved in the production of digital media. However, much existing research focuses on audience responses to AI disclosure, while less is known about how creators themselves understand disclosure decisions, interpret provenance information, and evaluate adoption conditions in their own work.This study investigates how visual content creators perceive AI-use disclosure and provenance mechanisms, using Content Credentials (C2PA) as a concrete example. The purpose is to generate creator-centred insights and translate them into actionable UX requirements for disclosureand provenance systems. The study is based on qualitative semi-structured interviews with ten visual content creators, including students, early-career creators, professionals, and freelancers working with illustration, graphic design, UX/UI, branding, marketing, and related visual practices. Participants discussed their creative workflows, AI use, disclosure attitudes, interpretation of current AI labels, and responses to a screenshot-based walkthrough of Content Credentials.The analysis resulted in six themes. Participants did not treat AI use as a binary disclosure trigger, instead, disclosure thresholds depended on creative contribution, final output involvement, and perceived consequences. Disclosure was valued as transparency and honesty, but also seen as a risk to perceived skill, originality, pricing, and professional value. Decisions varied acrosspublic, client-facing, educational, workplace, and platform governed contexts. Generic AI labelswere often considered too broad because they failed to communicate the type, stage, or degreeof AI involvement. Content Credentials were viewed as potentially useful trust infrastructure, but fragile due to metadata loss, bypassability, incomplete records, limited platform support, and possible misunderstanding. Adoption depended on workflow integration, visibility, creator control, and privacy boundaries.The study contributes empirically grounded UX requirements for creator-centred AI disclosure and provenance design. These requirements emphasise stage-specific disclosure, clearer distinctions between AI assistance and AI generation, layered information, context-sensitive disclosure profiles, provenance persistence, limitation communication, and creator control over public process information.