[manuscript] The Devil is in the Details [20 minute presentation] June 2023 While training design choices seem amoral, they interact to determine training (in)effectiveness, potentially harming/benefiting trainees and organizations. These moral implications intensify when training is administered at scale (e.g., e-training) and focuses on social issues like sexual harassment (hereafter, SH). In fact, research on SH training shows it can elicit trainees’ gender-based biases against content messengers. We suggest that one such bias, resulting from messenger gender-occupation incongruence and influencing training effectiveness, is lowered perceptions of the messenger’s integrity. We also investigate whether rich media will increase or decrease this perceived integrity penalty. Using an excerpt from real SH e-training and a sample (N = 210) consistent with the targeted training audience, we conducted a 2x2x2 relative comparison experiment (messenger gender x messenger occupation x media richness) and tested a moderated mediation model of the interactive effects of messenger gender-occupation incongruence and media richness on trainees’ perceptions of messenger integrity and training outcomes. Results suggest that trainees’ perceptions of messenger integrity decrease when the messenger’s gender is incongruent with their occupation, leading to worse outcomes in text-based training. These effects, however, are mitigated by increased media richness, providing support for media richness theory. Implications for researchers and practitioners are discussed.