Transparency in Generative AI Advertising: Brand Trust, Consumer Response, and Implications for Ethical Marketing Leadership

Abstract

The rapid integration of generative artificial intelligence into advertising has intensified questions of transparency, consumer trust, and ethical marketing leadership, particularly where firms must balance technological efficiency with responsible communication and informed consumer choice. Existing research has not sufficiently disentangled the effects of actual content authorship, artificial intelligence disclosure, and brand positioning within one experimental framework, leaving a gap in the business ethics literature on responsible artificial intelligence governance. This study examines how disclosure of generative artificial intelligence authorship affects brand trust and purchase intention in automotive advertising, whether these effects vary by content source and brand positioning, and whether human- and machine-generated copy differs in perceived linguistic quality when authorship is concealed. The analysis is based on a two-by-three-by-two between-subjects factorial experiment conducted in 2026 with 184 participants aged 18–45 recruited online in Poland, who viewed BMW or Tesla advertisements authored by a human, OpenAI GPT-4o, or Google Gemini 1.5 Pro under unlabelled or artificial-intelligence-labelled conditions. The empirical strategy combined reliability analysis, Type III factorial ordinary least-squares analysis of variance, heteroskedasticity-consistent robust inference, estimated marginal means, Holm-adjusted simple effects, difference-in-differences contrasts, and supplementary equivalence testing in Python 3.13.5. Under concealed-authorship conditions, perceived linguistic quality did not differ significantly across human, OpenAI GPT-4o, and Google Gemini 1.5 Pro copy, with F(2, 89) = 0.445 and p = 0.642, while equivalence tests did not establish formal equivalence. Artificial intelligence disclosure significantly reduced brand trust, with F(1, 172) = 9.11, p = 0.003, and ηp² = 0.050, and purchase intention, with F(1, 172) = 14.02, p < 0.001, and ηp² = 0.075; the author-by-disclosure interaction was also significant, with F(2, 172) = 3.20 and p = 0.043. The strongest disclosure penalties occurred for BMW advertisements generated by OpenAI GPT-4o (−1.275, Holm-adjusted p = 0.002) and Google Gemini 1.5 Pro (−1.111, Holm-adjusted p = 0.011), while the pooled genuine-artificial-intelligence difference-in-differences contrast showed a larger penalty for BMW than Tesla, with an estimate of −0.791, a 95% confidence interval from −1.491 to −0.091, and p = 0.027. These findings extend business ethics and ethical marketing leadership research by showing that transparent artificial intelligence disclosure can impose measurable trust and behavioral costs even when linguistic quality is not detectably inferior, supporting future work on brand–technology congruity, disclosure design, human oversight, and responsible artificial intelligence governance.