Women politicians face distinct abuse patterns across southern Europe study finds

The European Union has spent the past two years hardening its legal framework against online harassment. A directive on violence against women, adopted in 2024, criminalises cyberstalking and non-consensual intimate imagery across all twenty-seven member states. The Digital Services Act, fully applicable since February 2024, obliges very large platforms to assess and mitigate systemic risks to democratic discourse, explicitly including gender-based harm. Yet a new comparative study across five southern and western European countries suggests the policy response is still calibrated to the wrong problem.

What the data actually shows

Researchers applied Google's Perspective API, a toxicity classifier scoring text from zero to one, to replies sent to political candidates during election periods in France, Ireland, Italy, Portugal and Spain. Nearly 80% of the candidates monitored received at least one toxic reply. In raw numbers, women received more abusive messages than men. But the headline gender gap disappears once you control for reply volume: women posted less often, yet their accounts attracted more replies of every kind, toxic and benign alike.

The distinction sharpens when the analysis moves from volume to content. The team manually coded roughly 2,700 replies that the algorithm flagged with the highest toxicity probability, messages where nine out of ten human raters would agree on the label. Of those, 71% targeted women. The nature of the abuse diverged sharply. Male politicians were attacked for what they did: votes cast, expenses claimed, promises broken. Female politicians were attacked for who they were.

Sexualised silence as a political tactic

One in four women in the sample received comments of a sexual nature. The repertoire included explicit propositions, rape threats, speculation about their bodies, and demands that they exit public life. Researchers describe this as semiotic violence, language deployed not to contest ideas but to erase the speaker's legitimacy as a political actor. When abuse is identity-based rather than policy-based, the signal is not 'I disagree with you' but 'you do not belong here'.

The study also measured what it terms toxicity intensity: the number of distinct abusive elements packed into a single reply. A message aimed at a woman might combine a sexist slur, a sexual threat, a reference to mental instability and a demand to quit politics. Sixty-two percent of toxic replies to women contained two or more such elements. For men, 52% of toxic replies carried only a single element. Multiple simultaneous attacks amplify psychological impact and function as a coordinated signal that the recipient's presence is unwelcome.

Ideology compounds gender

Female candidates from left-leaning parties attracted disproportionately higher abuse volumes. The pattern mirrors findings from Brazilian election studies, suggesting an intersection where gender and progressive politics amplify targeting. Researchers caution that the five-country sample, all sharing Latin cultural roots and predominantly Catholic histories, may not generalise to northern or eastern Europe, though separate studies in Sweden, the Netherlands and New Zealand have documented the same gendered divergence: women do not always receive more abuse by volume, but what they receive is more personal, more sexualised, and more explicitly aimed at denying their right to hold office.

Where the legislation falls short

The EU directive on violence against women, formally Directive (EU) 2024/1385, establishes minimum criminal offences for cyber harassment, cyber stalking and the non-consensual sharing of intimate material. Member states had until June 2027 to transpose it. The directive text recognises that such acts disproportionately affect women and girls. Meanwhile, the Digital Services Act requires very large online platforms and search engines to conduct annual systemic risk assessments covering, among other things, 'gender-based violence' and 'effects on civic discourse and electoral processes'.

Both instruments treat abuse largely as a volume problem: detect more, remove faster, fine platforms that fail. The study suggests this misses the structural difference. A single sexually explicit threat carries different democratic weight than a corruption allegation, even if both score identically on a toxicity classifier. Moderation systems trained on aggregate toxicity scores will systematically under-detect the coordinated, multi-element attacks that characterise abuse against women.

The moderation gap

Platform trust-and-safety teams typically rely on keyword lists, hash-matching for known illegal content, and classifiers like Perspective API. These tools struggle with context: a slur reclaimed by a community, a threat veiled in irony, a pile-on where each individual reply falls below the toxicity threshold but the cumulative effect is silencing. Local language nuance compounds the problem, a sexist trope in Portuguese may not trigger a model trained predominantly on English data. The researchers argue for continuous, gender-disaggregated monitoring and for detection tools that understand conversational context, not just sentence-level toxicity.

Consequences beyond the screen

The democratic cost is measurable. Surveys of female parliamentarians in multiple European legislatures have found that online abuse correlates with reluctance to seek re-election, avoidance of certain policy portfolios, particularly gender equality, migration and reproductive rights, and reduced engagement with constituents on social media. When representatives self-censor, the policy agenda narrows. The study's authors note that the abuse operates as a de facto selection filter: women who might have run, don't; women who do run, modify their public presence; women who persist, expend disproportionate energy managing harassment rather than legislating.

What the next election cycle will test

The 2029 European Parliament elections will be the first full cycle where both the directive and the DSA's gendered risk obligations are fully in force. The Commission has signalled it will use the DSA's transparency database, which publishes platform risk assessments and mitigation reports, to compare how different services handle coordinated gendered campaigns. Civil society groups are preparing shadow reports. The real test is whether platforms can move from reactive takedowns to proactive detection of the multi-element, context-dependent attacks this study documents. If they cannot, the legal framework may simply produce better statistics on a problem that persists unchanged.