Kemi Chatbot
- Categories AI, Case Studies
- Date April 30, 2026
AI for Digital Safety: Evaluating the Kemi Chatbot for TFGBV Survivors in West Africa
A case study on Brain Builders Youth Development Initiative’s AI-powered WhatsApp chatbot for Technology-Facilitated Gender-Based Violence support.
Context & Problem Statement
Technology-Facilitated Gender-Based Violence (TFGBV) is an umbrella term for any form of abusive behaviour that targets someone because of their gender and is at least partly carried out using digital tools – including mobile phones, social media platforms, and messaging apps like WhatsApp. Cyberstalking, deepfakes, doxxing, online harassment, and non-consensual sharing of intimate images are among the rapidly evolving tactics perpetrators use.
In Nigeria, rapid growth in internet access over the past decade has made it one of the largest internet markets in the world. However, this digital expansion has also become a battleground where entrenched gender inequalities are replicated and amplified. As Hussain Taibat Aduragba, BBYDI’s TFGBV Programme Specialist, states: “As digital spaces become increasingly central to education, communication, commerce, and civic engagement, so too have they become battlegrounds where entrenched gender inequalities are replicated and amplified.”
Existing support systems for TFGBV survivors in West Africa face critical gaps: lack of confidential reporting channels, stigma preventing disclosure, limited multilingual resources, and few digital tools that meet survivors where they are. The problem spans multiple SDGs, primarily SDG 5 (Gender Equality) and SDG 16 (Peace, Justice and Strong Institutions).
Intervention Overview
Kemi is an AI-powered WhatsApp chatbot launched by Brain Builders Youth Development Initiative (BBYDI), a Girls Not Brides member based in Nigeria. Kemi offers confidential support to anyone who has experienced TFGBV.
Users can chat anonymously without fear of stigma or exposure.
Six locally spoken languages: English, French, Hausa, Yoruba, Igbo, and Pidgin.
No standalone app required – users simply send a message to a phone number.
What Kemi does: answers questions about different types of TFGBV, guides users through a health check to help them recognise abuse, signposts to services, and shares information about reporting cases – either to the online platform where the abuse occurred or to local authorities. The chatbot draws on a specially built West Africa-focused gender-based violence database.
Co-creation approach: BBYDI held 41 feedback and testing workshops across the region to ensure the solution was not just high-tech but high-trust. Survivors were directly involved in shaping the chatbot’s design, language, and responses.
Theory of Change
| Inputs | Activities | Outputs | Outcomes | Impact |
|---|---|---|---|---|
| AI system (Kemi) West Africa GBV database Survivor input & co-creation 41 feedback workshops Multilingual content | Chat-based guidance TFGBV health checks Awareness of abuse types Referral to services Reporting information | Confidential support available 24/7 Users understand TFGBV Users know how to report Increased help-seeking behaviour | Increased reporting of TFGBV Safer online behaviours Reduced stigma Improved access to justice pathways | Reduced harm from TFGBV Strengthened digital safety for women and girls Progress toward SDG 5 |
Evaluation Objectives
- Assess the effectiveness of Kemi’s chatbot support in increasing help-seeking behaviour.
- Measure user engagement, trust, and satisfaction across different language groups.
- Evaluate accessibility and inclusion, identifying who is using Kemi and who is left out.
- Analyse contribution to reporting behaviour and referral uptake.
- Examine ethical safeguards and data protection from a survivor-centered perspective.
Key Evaluation Questions
Does Kemi improve help-seeking behaviour among TFGBV survivors?
Does the chatbot respond to survivors’ real needs as identified through co-creation workshops?
Who is using Kemi – and who is being left out?
Is user data protected? Is anonymity maintained?
Can Kemi scale across other countries in West Africa?
Methodology
A mixed-methods evaluation approach is recommended for Kemi, combining quantitative analytics and qualitative user feedback.
- Chatbot analytics: Conversation logs (anonymised), user flow data, drop-off rates, language selection patterns.
- User feedback surveys: Embedded within chatbot flow to capture satisfaction, trust, and perceived usefulness.
- Co-creation workshop outputs: Qualitative insights from 41 pre-launch workshops across the region.
- Referral tracking: Uptake of signposted services and reporting mechanisms (where ethically permissible).
- Digital divide analysis: Assess smartphone/internet access across user demographics.
Note for M&E professionals: AI-powered interventions like Kemi offer a unique opportunity for real-time monitoring. Chatbot analytics can provide continuous data on user needs, help-seeking patterns, and language preferences, enabling rapid iteration and improvement.
Key Findings
- High-trust design drives adoption: 41 co-creation workshops ensured survivors shaped the chatbot, building confidence that Kemi understands their context and needs.
- WhatsApp accessibility is a key enabler: By using a widely adopted messaging app, Kemi removes barriers associated with standalone apps – no download required, works on basic smartphones.
- Multilingual support covers major West African languages: Six languages address Nigeria’s linguistic diversity, increasing reach and relevance.
- Health check feature empowers users to recognise abuse: Kemi’s guided questions (e.g., “Are you experiencing repeated unwanted contact or harassment online?”) help survivors name their experience.
- Technology as a double-edged sword: The same digital access that enables TFGBV also enables support – but requires active ethical management.
Challenges & Limitations
- Digital divide: Users without smartphones or reliable internet access cannot reach Kemi.
- Privacy and data risks: WhatsApp conversations, while encrypted, still require careful handling of sensitive survivor information.
- AI limitations: Chatbots cannot fully understand complex trauma, may misinterpret language nuances, and cannot replace human counsellors.
- Lack of integration with offline services: Referral pathways depend on availability of local support services, which may be limited in some areas.
- Trust in automation: Some survivors may prefer human interaction, and the chatbot may not be appropriate for all users or all stages of their journey.
Lessons Learned
- Co-creation is non-negotiable: 41 feedback workshops ensured Kemi wasn’t just high-tech but high-trust. Survivor input must be embedded from the start.
- Platform choice drives adoption: WhatsApp’s widespread use in West Africa made Kemi accessible without requiring new digital literacy.
- AI works best as a triage tool, not a replacement: Chatbots can guide, inform, and signpost – but must connect survivors to human support systems for complex cases.
- Ethical safeguards are foundational: In GBV contexts, data protection, anonymity, and informed consent are not optional – they are core features.
Recommendations
- Strengthen referral pathways: Map and integrate with local offline support services (counsellors, legal aid, shelters) across West Africa.
- Expand language localisation: Include more indigenous languages beyond the current six, and adapt content for different cultural contexts.
- Improve data protection frameworks: Implement clear policies on data retention, anonymisation, and user control over conversation history.
- Scale to other regions: The Kemi model can be adapted for other countries in West and Central Africa, with contextualisation of language and referral pathways.
- Establish a human backup system: Provide a clear escalation pathway to trained counsellors for users who need more than a chatbot can offer.
Implications for M&E Practice
Chatbots generate real-time analytics that can inform programme adaptation and early warning systems.
M&E for AI tools should embed continuous feedback loops, not just periodic evaluations.
Evaluators must prioritise survivor safety, confidentiality, and do-no-harm principles.
Co-creation doesn’t stop at design – survivors should inform evaluation questions and methods.
Key Takeaways
- AI can enhance survivor support if built on a foundation of trust and co-creation with affected communities – demonstrated through 41 feedback workshops.
- Delivery channel matters: WhatsApp’s ubiquity drives adoption, but integration with offline services remains essential.
- Data from chatbot interactions provides real-time insights for M&E professionals, enabling rapid iteration and evidence-based improvement.
- Ethical risks (privacy, bias, misinterpretation) must be actively managed, not assumed away, in any AI-for-GBV intervention.
References & Sources
- Girls Not Brides. (2025). Meet Kemi, the WhatsApp chatbot supporting survivors in West Africa. Retrieved from girlsnotbrides.org
- Brain Builders Youth Development Initiative (BBYDI). TFGBV Programme Reports and Documentation.
- Unicef & HIAS. (2021). Report on migrant, displaced and refugee girls in Latin America and the Caribbean.
- Girls Not Brides. (2021). Impact of Covid-19 on efforts to end child marriage.
© EvalCommunity — Case study based on Girls Not Brides article (December 2025) and BBYDI documentation. For professional M&E training on AI, gender equality, and digital safety, visit EvalCommunity Academy.
EvalCommunity does not claim ownership of the original programme or content. Readers are encouraged to consult original sources for complete context.
The courses and articles have been developed by an experienced team of evaluators and software developers under the guidance of Fation Luli. The EvalCommunity Academy combines practical expertise in Monitoring & Evaluation with cutting-edge AI technologies to provide high-quality, accessible learning experiences for professionals around the world.