Visual Stereotypes in AI: A Content Analysis of AI-Generated Imagery Representing Traditional Nigerian Gender Roles

Background: The rapid integration of Generative Artificial Intelligence (GenAI) into visual media has revolutionised digital content creation but has also raised critical concerns about the replication of societal biases. While GenAI tools offer unprecedented opportunities for cultural storytelling, they often function as algorithmic mirrors that reflect and amplify historical prejudices. In the context of the global South, there is a growing risk that these technologies facilitate a form of digital colonisation by reducing complex identities to Western-centric tropes. Objectives: This study investigated the manifestation of visual stereotypes in AI-generated imagery, with a focus on traditional Nigerian gender roles. It aimed to identify predominant gender tropes, assess the extent of vocational and domestic stereotyping, and evaluate the degree of cultural authenticity and ethnic flattening in the outputs of leading generative models. Methodology: A quantitative content analysis was conducted on a purposive sample of 200 images generated by Midjourney (v6.0) and DALL-E 3. Standardised, neutral prompts were used to generate depictions of traditional Nigerian men and women. The images were coded across dimensions of setting, labour type, power dynamics, and cultural attire. A Chi-square test of independence was performed to determine the statistical significance of gender-based role assignments. Results: The findings revealed a persistent reliance on traditional binaries. Female subjects were predominantly in domestic settings (72%) and assigned to reproductive labour, while male subjects were in public or vocational spheres (68%). Statistically significant disparities were also found in visual gaze and social positioning, with men depicted as authoritative figures and women as passive participants. Furthermore, a qualitative assessment highlighted a high frequency of ethnic flattening and the persistence of the primitive trope. Conclusion: The study concludes that current GenAI models reinforce restrictive, patriarchal, and colonial-era representations of Nigerian identity. These findings underscore the urgent need to decolonise AI datasets and integrate indigenous epistemologies into model development to ensure more inclusive and culturally accurate digital representations.