Willing everywhere but in health: a within-person national survey of artificial intelligence acceptance among 11 013 Nigerian adults
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Introduction
Artificial intelligence is entering African primary care partly on the premise that populations underserved by clinicians will accept an algorithm in a clinician’s place, a premise rarely tested where it matters. We measured how far Nigerian adults separate health from every other use of AI in their lives.
Methods
Cross-sectional national survey of 11 013 adults across Nigeria’s 36 states and Federal Capital Territory, July–October 2025. About 60% were recruited face to face, with translation where needed, so neither literacy nor English was required to take part. Each rated willingness to use AI in nine everyday situations on a 1–5 scale, one being health advice when no doctor was available. The primary outcome was a within-person difference: health minus the respondent’s mean of the other eight. Associations were estimated by ordinary least squares with cluster-robust (CR2) standard errors by state.
Results
Health finished last of the nine (mean 3.10), below promoting one’s own political party (3.40). The within-person gap was −0.568 (95% CI −0.594 to −0.543; d z −0.417), widening to −0.904 among the 6 924 who distinguished between domains. Keeping a clinician in view did not relieve it: 22.9% would trust AI to help doctors diagnose against 25.5% who would use it to check their own symptoms. Among barriers respondents named, the gap was wider for language (β −0.427, 95% CI −0.718 to −0.136) and unfriendly staff (−0.411, −0.616 to −0.207), narrower for missing equipment (+0.231, +0.040 to +0.423) and narrower among fluent English speakers (+0.386 per SD, +0.130 to +0.643).
Conclusion
Nigerian adults accept AI across their lives and withhold it from their health, most firmly where care has been unaffordable, unintelligible or unkind. No configuration tested commanded majority willingness, including AI presented as assisting a clinician; willingness belongs alongside connectivity and data in deployment planning, not assumed.
Key questions
What is already known on this topic
-
Health systems in low- and middle-income countries are being encouraged to adopt patient-facing AI, including symptom checkers positioned as a first point of contact without provider oversight, on the reasoning that people will turn to AI where access to care is poor.
-
The best-identified experimental default in lay judgment is algorithm appreciation rather than aversion; resistance to medical AI is compensatory and can be bought off by demonstrated accuracy.
-
Patient-attitude evidence is dominated by high-income settings and hospital samples; reviews and strategies that enumerate barriers to health AI in Africa do not count population willingness among them.
What this study adds
-
In a national sample of 11 013 Nigerian adults, health was the least acceptable of nine uses of AI, below promoting one’s own political party, with the comparison drawn within each respondent so that scale-use habits cancel.
-
The discount was not explained by stakes, by inattentive answering, or by keeping a clinician in the loop, and it was not closed by digital capability.
-
It was largest among respondents whose stated grievance with the health system was language or disrespect, and among the uninsured, and smallest among those whose grievance was missing equipment: the gap tracks how people were treated rather than what the clinic lacked.
How this study might affect research, practice or policy
-
Not one of the four framings tested, including AI presented as assisting a clinician, reached majority willingness, so willingness is a constraint to be measured before deployment rather than a barrier to be educated away.
-
Willingness is cheap to measure and belongs in African health-AI strategy alongside connectivity, data and governance, and in these data the sharpest predictor of it is language.
