Working together for women's safety online

International partnerships grounded in trust, local expertise and collective decision-making built credible evidence of people’s lived experiences of online misogyny and gender-based violence in Ethiopia. Now armed with validated data, a contextualised lexicon and policy-ready recommendations, partners are shaping policy and influencing social media platform providers to create safer digital spaces for women and girls.

How the impact happens 

“We started off by simply approaching organisations to learn more about their work and challenges,” says Felicity Mulford from the Centre for Information Resilience (CIR). Those early discussions revealed a common concern: online gender-based violence was widespread, yet the absence of evidence in many countries meant women’s experiences were routinely dismissed as anecdotal. “There was no data for the Ethiopia context – just personal testimonies that were often disregarded as one-offs,” she explains.  

This harm appears in many forms on social media – from coordinated harassment to threats – often silencing women and girls, damaging their reputations and pushing them out of public online spaces. Initial conversations evolved into workshops and roundtables and partners moved from being contributors to trusted co-designers. But without a fuller picture of the extent of the abuse, campaigning and advocacy would be brushed off. So, CIR teamed up with The University of Manchester and local language experts in Ethiopia to conduct a systematic quantitative study that analysed hate-containing posts on social media platforms, to build that evidence base. 

The work began by listening before acting. It was essential to properly define the study’s scope, definitions and agree on which platforms to investigate. Ethiopian civil society partners helped shape which questions were asked and decide what evidence would be genuinely useful. For Riza Batista-Navarro, an academic specialising in text mining who led the computational aspects of the project, this required her team to recognise their limitations even from the very beginning of the project. “I don’t speak the native languages... so at the beginning I was like, is this going to work out? Hate speech is very much based on culture,” she says. That uncertainty pushed the team to slow down and listen carefully. 

Talking through tension 

There were moments of tension. Even among native speakers, the definitions of sarcasm and irony – and how they relate to hate speech – were debated. Rather than smoothing over disagreement, the team treated it as part of the process. Nuhu Ibrahim, a PhD student specialising in AI, was working on the project and observed the remarkable receptiveness of the team and their overriding desire to learn from each other: “Everyone was open-minded and willing to contribute from their own perspective and that’s the reason why the project was successful.” 

By sharing decisions and adapting together, the partnership produced evidence that communities trusted and could use to campaign for safer online spaces for women and girls. Insights were not delivered to people – they were built with them. 

Moving forward 

The project team is now extending the same co-design approach to new platforms such as TikTok and YouTube, where online abuse increasingly shapes public debate. Rather than treating this as a purely technical challenge, the focus remains on adapting methods with care. As Riza reflects: “What you consider as hate speech and what I consider as hate speech, can vary depending on the cultures we come from, so you have to listen.” This principle continues to guide how multilingual word lists and coding frameworks are developed for new case studies, working closely with local experts to ensure cultural accuracy. 

Follow-on research addresses a key gap identified by the project. An Africa Futures PhD studentship at The University of Manchester is focused on improving how AI tools handle languages lacking large-scale digital datasets. For Felicity, meanwhile, long-term success will be measured by use, not publication: “We now have data that people can actually take forward. Something authorities should find convincing and reliable and evidence that digital platform providers simply cannot ignore.” 

Looking ahead, the ambition is clear: these co-created findings are now informing advocacy and underpins national policy debate. Thanks to a sustained partnership, the collaboration now has a labelled dataset and multilingual lexicon that underpin 34 national policy recommendations. Through evidence-led activism, women and girls are fighting back to feel safer online.