Disclosing the Artificial: How AI-Image Labels Shape First Impressions and Trust on Institutional Websites : First Impressions and Deliberate Trust Reasoning, using domstol.se as an illustrative case of a high-trust judicial context.

Sammanfattning

AI-image disclosure is an emerging user experience challenge, where transparency requirements meet users’ expectations of credibility, professionalism, and human responsibility. While emerging regulation such as the EU AI Act requires certain AI-generated and AI-manipulated content to be disclosed, less is known about how users notice, interpret, and reason about such disclosure cues during first exposure. This is especially relevant in high-trust institutional contexts, where users expect information to be reliable, serious, and accountable. This thesis explores how users reason about AI-image disclosure in the initial viewport of an institutional website, using Domstol.se as a representative case of a Swedish judicial context. The study used a qualitative research design based on a high-fidelity prototype of an institutional webpage containing an AI-image disclosure label. Data collection was structured in two phases. In the first phase, ten participants viewed the prototype during a brief 10-second first-impression exposure and provided immediate ratings of trust, credibility, and visual appeal. In the second phase, participants took part in semi-structured interviews where they reflected more deliberately on what they noticed, how they interpreted the disclosure, and how it affected their reasoning about the website. The data were analysed using reflexive thematic analysis. The findings show that AI disclosure cannot be understood only as the presence of a visible label. Although the disclosure was placed in the initial viewport and combined both icon and text, it was not always noticed during brief exposure. When the label was noticed, it was not always interpreted as intended, as some participants were uncertain whether it referred only to the hero image or to the wider webpage. The findings also show that participants’ first impressions were strongly shaped by familiar institutional cues, such as official visual identity, structured layout, and professional appearance. Once the AI disclosure was processed, it sometimes introduced questions about authenticity, human effort, accountability, and institutional appropriateness. Visual appeal often remained stable, while trust and credibility became more conditional. The thesis contributes user-centred insight into AI disclosure design by showing that meaningful transparency requires more than formal visibility. In high-trust institutional contexts, disclosure labels need to be noticeable, specific, understandable, and context-sensitive in order to support trust rather than create ambiguity.