evidoria

← Back to browse

Good practice Imported

MAPEMA — Kenya's AI-Assisted Hate-Speech and Disinformation Monitoring for the 2022 Election

Kenya · Nairobi · See the Kenya profile · See the Nairobi profile

Evidence: Descriptive / self-reported Top 83% 40/100 · Ask Evidence Copilot about this practice

Ahead of Kenya's 2022 election, NCIC partnered with Code for Africa on an AI-assisted monitoring effort that flagged over 550,000 toxic Facebook posts and 800+ hate-speech cases, though the AI tooling was built and run by an NGO consortium, not government itself.

550,000+ posts
Toxic Facebook posts flagged (2022 election period)
800+ cases
Hate-speech cases referred to platforms (2022 election period)

Details

Promoter
National Cohesion and Integration Commission (NCIC) and UWIANO Platform for Peace, with Code for Africa, UNDP and OHCHR
Period
2022
Keywords
Electoral integrity, Peacebuilding, Civil society

Context

Ahead of Kenya's August 2022 general election, the National Cohesion and Integration Commission (NCIC) — Kenya's statutory body for social cohesion — worked with the government-run UWIANO Platform for Peace, UNDP, OHCHR and the civil-society organisation Code for Africa to counter hate speech and disinformation on Facebook, X/Twitter and Instagram in English, Swahili and Sheng.

Objectives

To combine Code for Africa's CivicSignal media-monitoring stack with an AI classifier (reportedly built on Primer.ai) to flag toxic and inciteful content in near-real time during the election period.

Activities

The AI tooling was built and operated by the Code for Africa-led consortium rather than by NCIC itself, with NCIC and UWIANO acting as government partners receiving the monitoring outputs rather than technical owners of the system.

Results

According to Code for Africa's own reporting, the effort identified more than 550,000 toxic Facebook posts and referred over 800 hate-speech cases to social media platforms for action during the election period. The UN Sustainable Development Group and the Observer Research Foundation later cited the initiative as an example of AI-assisted election monitoring in Africa.

Conclusions

The headline figures are self-reported by the implementing NGO and have not been independently audited; no precision or false-positive rate has been published, and there is no confirmed evidence the same AI stack continued operating after the 2022 cycle. Human Rights Watch and ARTICLE 19 have separately raised concerns that NCIC's hate-speech enforcement powers have, at times, risked chilling legitimate political speech.

Implementation
Indicative cost
Low (< €50k) — No cost figures published; inferred as low-cost given the described scope (a time-bound, single-election social-media monitoring effort run by an NGO consortium) — inferred from programme description, not stated.
Time to results
Short (< 1 year) — Monitoring effort was tied to the August 2022 election period; no confirmed evidence it continued as a standing capability afterward.
Staffing & skills
NGO consortium (Code for Africa) technical team building and operating the CivicSignal stack and AI classifier, NCIC and UWIANO Platform for Peace as government partners receiving monitoring outputs, UNDP and OHCHR as supporting international partners

Conditions for success

  • Partnership pairing a government cohesion body with an independent civil-society technical implementer
  • Multi-language classifier covering English, Swahili and Sheng to match local social-media discourse

Common failure modes

  • Self-reported metrics with no independent audit or published precision/false-positive rate
  • No confirmed evidence the AI stack continued operating after the 2022 election cycle
  • NCIC's own hate-speech enforcement powers have separately drawn concern (HRW, ARTICLE 19) for risking a chilling effect on legitimate political speech

Where it fits

Governance type
national government body + civil-society consortium
Scale
national (Kenya, 2022 general election)
Income level
lower-middle-income

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Similar practices you may find useful