AI and Elections: Chatbots Still Unreliable, Warns AfricTivistes-DRI Study

21 août, 2025

In an increasingly digital electoral landscape, where artificial intelligence shapes access to information, AfricTivistes and Democracy Reporting International (DRI) today publish a study assessing the performance of chatbots in the face of democratic challenges.

Conducted in April 2025, the research focused on five generative AI tools (Copilot, Gemini, ChatGPT-4o, ChatGPT-4omini, and Claude) and their ability to answer 14 questions in French and Wolof concerning Senegal’s legislative elections held on 17 November 2024. The analysis was based on three key criteria: accuracy, consistency, and completeness of responses.

The findings are striking:
🔻 ChatGPT-4o and Claude performed relatively well but still showed significant limitations.
⚠️ Copilot and Gemini frequently provided incomplete, vague, or even outdated answers.
🚫 ChatGPT-4omini showed the poorest performance, with frequent errors.

Worse still, some chatbots provided incorrect information with high levels of confidence, posing a serious risk of disinformation—especially during electoral periods. None were able to integrate recent developments in Senegal’s political context, revealing a structural inability to update in real time.

Beyond this diagnosis, the study calls for heightened caution in the use of generative AI and underscores the urgent need to strengthen digital literacy, algorithmic transparency, and the accountability of tech platforms.

As many African countries enter electoral seasons in 2025, this study stands as a civic warning and a reminder not to mistake automation for reliability.

AI and Elections: Chatbots Still Unreliable, Warns AfricTivistes-DRI Study Télécharger