A recent evaluation of a Microsoft-backed artificial intelligence chatbot, formerly known as Bing and now operating under the Copilot branding, reveals notable gaps in accuracy when discussing electoral events in Europe. The assessment was carried out by AlgorithmWatch, a human rights research organization, and its findings highlight a pattern of incorrect or misleading information that could shape readers’ understanding of political processes. The partnership produced a detailed report that documents how the chatbot handles inquiries about elections in the Bavarian and Hesse regions of Germany, as well as in Switzerland. The key takeaway is that a significant portion of the bot’s responses contained factual errors, and in several instances the system avoided offering clarifications that would help users interpret the information correctly. Overall, only a minority of the chatbot’s statements were factually reliable in the examined scenarios.
The study shows that the information returned by the Copilot model sometimes included fabrications such as invented debates, erroneous election dates, and incorrect polling data. In some cases, the chatbot referenced candidates who were not part of the electoral process, painting a misleading picture of the electoral landscape. The researchers noted that even when the model was fed data from a single source, the numbers and conclusions presented in its replies did not always align with the original material. At times, the bot even reordered parties relative to the source data, which could confuse readers who rely on source accuracy to gauge political sentiment and potential outcomes.
AlgorithmWatch emphasized that the selected elections were among the first major electoral events conducted in the region since Copilot’s deployment, enabling a clearer comparison of how the system performs across languages and local contexts. The project examined responses in German, English, and French to identify linguistic nuances that might influence accuracy and interpretation. This multilingual dimension is important because users in Canada and the United States frequently interact with multilingual content about international events, and language choices can affect the reliability of automated answers.
In reacting to the findings, Microsoft acknowledged the gaps observed in the study and outlined intentions to enhance the artificial intelligence system’s reliability ahead of forthcoming electoral cycles in North America and beyond. The company stated that improvements would focus on maintaining factual integrity, improving transparency about sources, and offering clearer caveats when information may be uncertain. The objective is to help users obtain verifiable information that supports informed decision-making, especially during politically sensitive periods when misinformation can spread rapidly.
The report also notes broader concerns about the potential risks associated with artificial intelligence in financial and political contexts. While the study centers on election data, it underscores the need for ongoing oversight and refinement of AI systems that generate public-facing information. The aim is to reduce the likelihood that automated responses mislead readers or propagate errors that could influence opinions, choices, or market expectations. Industry observers continue to call for robust evaluation mechanisms, better source attribution, and safeguards that balance speed and accuracy in automated answers.