EXECUTIVE SUMMARY

This report presents the findings of Build Up’s social media monitoring and qualitative analysis of Somalia’s online electoral and democratic dialogue, conducted between 01 September 2025 and 31 March 2026. Drawing on data from Facebook, TikTok, X, and mainstream media, it provides an integrated account of the dominant political narratives circulating online, the reception of those narratives in the comment sections of the most-engaged posts, and the harmful content and behaviours that accompanied them.

Somalia’s 2026 electoral moment unfolds within a contested federal settlement that remains unsettled in both principle and practice. The review period sits within a renewed phase of that contest, encompassing the constitutional amendment and signing of March 2026, the December 2025 Benadir local elections, the dispute over the South West State, and competing claims over the legitimacy of the government’s mandate beyond 15 May 2026. In this context, social media has become a central arena in which citizens, the diaspora, political actors and regional administrations interpret events, contest legitimacy, mobilise support and advance competing visions of the country’s democratic future.

Across approximately 156,000 pieces of content collected during the period (1 September 2025 to 31 March 2026), of which 29,550 posts were classified into topic areas, the analysis tracked nine narratives spanning the election model, government-opposition dialogue, the constitutional amendment, the FGS mandate, and public land evictions. For each narrative, the team read a sample of the most-engaged comment sections in Somali to assess public reception and to identify and categorise harmful content. The data show that digital harms were not isolated incidents but recurring features of online political discourse, concentrated on the highest-engagement posts and around offline flashpoints. Notably, Facebook and TikTok carried the overwhelming majority of the harmful content observed.

The key findings and recommendations that follow are intended as a practical resource for civil society, media actors, electoral and governing institutions, political actors and international partners working to safeguard civic space, information integrity and a peaceful electoral process in Somalia.

Key Findings

Key Finding 1: The model of election was the most-discussed topic in the dataset, and online opinion split into two opposed narratives, a high-volume, celebratory case for direct elections / qof iyo cod (1.35k posts) and a smaller, more combative case that the vote was a one-party, militarised “qof iyo qori” exercise (276 posts). Harmful content was comparatively limited, surfacing mainly as clan-based smears against opponents of the direct vote. Key Finding 2: The dialogue process was widely framed as stalled or performative, and international partners were cast in two opposing lights, legitimate mediators versus illegitimate “meddlers”, with the most-engaged comments attacking Somali leaders for appealing to foreign envoys. The harmful fringe was small but severe, including antisemitic conspiracy and an explicit call for Iran to strike Somali cities. Key Finding 3: The constitutional amendment was the most polarising topic in the dataset, dividing cleanly into pro and anti-amendment camps. The allocation of upper-house (Senate) seats was the main channel for clan-identity hostility, alongside foreign-conspiracy and antisemitic smears, a distinctive pro-amendment framing of critics as “Al-Shabaab-minded”, and viral disinformation that each lawmaker was paid a US$400,000 bribe. Key Finding 4: The FGS mandate debate produced the sharpest incitement observed in the dataset. A pro-government strand called for critics to be jailed or even killed, while an anti-government strand mobilised for 15 May confrontation (“ready with stones”, “take up arms”); the pro-mandate side also used racialised clan-hierarchy hate speech to delegitimise opponents. Key Finding 5: Evictions were the highest-volume single topic and the most hostile comment environment in the dataset, directed mainly at the President and federal leadership. Clan-based hostility was the largest category of harmful content, markedly more visible in the comments than in the posts themselves, and included direct incitement to seize or burn installations and demands that leaders “be killed”. Key Finding 6: Because harmful authorship is so diffuse, the most frequent accounts are named illustratively rather than as evidence of orchestration: no small group drives the harmful content, which instead surfaces across a very large number of one-off contributors. The value of listing them is precisely to show that even the ‘top’ harmful accounts contributed only a handful of comments each, a markedly different pattern from the cross-topic connectors above, and one reason the harm is better understood as a broad cultural register than as a coordinated campaign.

Recommendations

The findings of this report point to the need for practical, coordinated solutions ahead of Somalia’s 2026 democratic and electoral milestones. Authorities should protect peaceful dissent, respond lawfully, and communicate transparently on the constitutional, electoral and mandate processes. Electoral institutions should embed monitoring of hate speech, incitement and disinformation into integrity planning, while platforms, especially Facebook and TikTok, strengthen Somali-language moderation and prioritise viral content during sensitive periods.

Civil society and media should counter pervasive clan-identity hostility through de-escalation, rapid verification and victim support, while political actors disavow incitement and clan-baiting. International partners should sustain independent monitoring, fund media literacy, and encourage dialogue while staying alert to anti-foreigner narratives. Build Up, SPA and PDRC call for a long-term commitment from all actors, and for expanded regional analysis to reduce Mogadishu overrepresentation, to build a resilient, rights-respecting information environment.

A full analysis of the 1.6million comments collected could only be done if an automated labelling using Natural Language Processing (NLP) techniques in Somali could be used. However, currently we have no model that has been trained to identify harmful content in the categories used for this analysis, in Somali language. This is a limitation that Build Up is actively addressing through its programming.