On 22 June 2026, UN Women published figures on gender inequality in artificial intelligence. A study of 133 AI systems shows that 44 per cent exhibit gender bias, and 26 per cent display both gender and racial bias. Large language models still associate women with home and family, and men with career and salary. In sentence-completion tasks, around 20 per cent of responses from these models reflected sexist views.
Only 51 per cent of marketers use human oversight to test AI-generated content before release. Because AI systems draw on historical data, they inherit existing inequalities. Of 138 countries analysed, 24 mention gender in their national AI strategy, but only 18 set out substantive rules. Meanwhile, women make up 30 per cent of the global AI workforce.
Impact on public communication
These patterns affect non-profits and government bodies directly. AI models generate summaries and responses for citizens using existing online sources. If those underlying models are biased, stories and data from specific groups risk being misrepresented or left out entirely.
If you work for a local council and use generative AI for public information, software running on biased models can create stereotypical texts unnoticed. Publishing clear, inclusive information on your own website helps.
Ensuring inclusive content
AI systems can reinforce stereotypes, but deliberate configuration can also help detect them. You can manage risks in your organisation’s communication by keeping human oversight in place:
- Manually check all AI-generated text and images for stereotypical assumptions before publishing.
- Ensure varied representation across the examples and stories on your website.
- Involve subject-matter experts to review AI output on sensitive societal topics.
Source: UN Women
