Generative AI and migration narratives: algorithmic bias and discursive patterns in African migration to Spain

Abstract

The use of generative language models such as GPT-4o introduces new challenges for the critical analysis of migration narratives. This article examines how such models discursively construct narratives about African migration to Spain, based on a corpus of 200 automatically generated narratives derived from 50 systematically varied prompts. Drawing on tools from critical discourse studies, the methodology combines lexical analysis (Voyant Tools) with qualitative thematic coding (NVivo) to identify dominant narrative patterns. The results identify four recurring archetypes: the victim migrant, the resilient migrant, the NGO saviour, and the bureaucratic state. These narratives are structured around compassionate-paternalistic tones and utilitarian logics, framing migrant agency within normative expectations of integration, merit, and gratitude. Prompt formulation operates as a discursive activation mechanism that delimits the range of meanings and reinforces dominant narrative patterns while silencing hybrid or dissenting voices. The findings suggest that generative AI systems do not merely reflect existing social imaginaries, but actively reorder and amplify them through algorithmic discursive logics. The study highlights the need for ethical and methodological frameworks in AI-generated representations of migration and proposes strategies for enhancing linguistic, epistemic, and representational diversity. These findings contribute to a broader critical reflection on the relationship between discourse, artificial intelligence, and mediated narratives of migration.