GENERATIVE AI AND ELECTORAL INTEGRITY IN NIGERIA'S 2027 ELECTIONS: TRUST, MANIPULATION, AND DEMOCRATIC RESILIENCE IN THE GLOBAL SOUTH

Commentary on generative artificial intelligence and elections predicts catastrophe, and the prediction has repeatedly outrun the evidence. This article rejects both the panic narrative and its complacent mirror image, arguing that synthetic content harms elections through a conditional trust pathway rather than a technological main effect. We develop and test this position for Nigeria’s 2027 general elections, where the 2023 cycle already produced documented synthetic-media incidents inside a low-trust institutional environment that Western research does not model. A five-stage conceptual model—generative content, trust assessment, information verification, political judgement, electoral behaviour—was tested through a Qualtrics, six-arm survey experiment (N = 2,400) crossing authenticity, disclosure, and modality. Three results invert the panic narrative. First, synthetic exposure damaged election-process trust substantially (d ≈ 0.58) while leaving candidate evaluations essentially unmoved, confirming that the operative channel is institutional trust, not persuasion. Second, a generic fabrication warning attached to authentic content depressed its credibility (d ≈ 0.49) and lowered process trust (d ≈ 0.42)—a scepticism tax that served the liar’s dividend without any fake present. Third, synthetic audio evaded detection far more effectively than synthetic video (48% vs. 66% correctly identified), exposing the video-centred bias of the existing literature against the audio-dominant Global South threat surface. Disclosure labels partially restored scepticism toward fakes but did not repair trust. The article contributes a Global South-grounded, condition-based theory of synthetic-media harm and a policy framework that targets trust infrastructure rather than content alone.