Analysis of 1,083 verifiable citations across ChatGPT, Gemini, DeepSeek and Perplexity finds nearly 7 in 10 come from UN or multilateral organisations
NAIROBI, Kenya, August 2026 –Only 5.5% of citations used by four major AI platforms to answer questions about humanitarian crises in Africa came from organisations based in the countries in crisis, a new 10-country study released on Monday 24th,August 2026 has found.
Nearly seven in 10 citations, 68.8%, came from UN or other multilateral organisations.
The African Humanitarian AI Citation Index (AHACI) examined sources cited by ChatGPT, Gemini, DeepSeek and Perplexity when answering questions about Burkina Faso, Chad, Democratic Republic of the Congo, Ethiopia, Kenya, Mozambique, Nigeria, Somalia, South Sudan and Sudan. The analysis covered 1,083 verifiable citations.
“The crisis may be local, but the visible authority explaining it to AI users is usually international,” said Hezron Ochiel, founder of HezronInsights.com and lead researcher.
The findings raise a critical question as AI becomes a first stop for information: whose knowledge becomes visible when AI explains a crisis?
Kenya stands apart
National differences were stark. Of the 60 country-local citations identified across the dataset, 31 came from Kenyan sources – more than half of all local citations. Sources included media organisations, humanitarian institutions and public agencies. Nigeria also recorded relatively stronger local visibility.
The Democratic Republic of the Congo showed the opposite pattern. Of 100 verifiable citations on the DRC, none came from a source based in the country. Sudan also recorded no country-local citation, while Somalia recorded only one.
The study does not establish why the differences occurred. It points to possible factors including language, digital publishing, search visibility, institutional authority and how easily African organisations make their information discoverable to AI systems.
Five organisations dominate AI attention
Citation visibility was highly concentrated. The five most-cited knowledge producers accounted for 40.3% of all verifiable citations. The top 10 accounted for 58.8%. UNICEF was the most cited, followed by UNHCR, OCHA/Humanitarian Country Teams, OCHA and the International Organization for Migration (IOM).
Even when local evidence appeared, it rarely stood alone. Nearly 85% of AI responses containing at least one country-local source also cited a UN or multilateral organisation. Only one response relied exclusively on country-local sources.
“Local organisations are producing useful knowledge every day. We need to make sure AI can find that knowledge and show it when people search for answers,” Ochiel said.
A separate analysis of ChatGPT alone – the only platform with complete source-level data for all 90 questions – showed a similar pattern: 7.2% country-local citations versus 70.9% UN/multilateral.
A new question for localisation
Humanitarian localisation has traditionally focused on who receives funding, who implements programmes and who participates in decisions.
The findings introduce another dimension: who gets to be visible when AI becomes an entry point to information about a crisis?
As AI answers increasingly shape what journalists, researchers, policymakers, donors and the public encounter first, citation visibility can influence which organisations are repeatedly presented as authoritative.
The study does not prove bias against African sources, nor does it assess how many relevant local sources were available per question. It measures the sources users were visibly shown.
About the study
AHACI’s first batch contained 100 independently randomised questions covering 10 countries and 10 humanitarian themes. Ten were used to refine the protocol.
The final analysis used 90 questions across the four platforms, generating 360 AI responses on conflict, displacement, food insecurity, health emergencies, water and sanitation, climate shocks, humanitarian access, protection, livelihoods and humanitarian response.
All questions were asked in English, a key limitation for Francophone and Lusophone countries. The manuscript is prepared for journal submission and has not yet been peer-reviewed.
