The use of chatbots, such as OpenAI’s ChatGPT 3.5 and 4.0, Google’s Gemini, and Microsoft’s Copilot,  to inform voters ahead of the 2024 European elections is leading to unintentional disinformation. This is the conclusion reached by a recent study entitled “Are Chatbots Misinforming Us About the European Elections? Yes” published in April 2024 by Democracy Reporting International, a non-profit organisation based in Berlin. The study methodology involved asking four chatbots ten questions in ten EU languages about electoral information in ten  EU member states.

Although the chatbots appear to have been designed to avoid partisan responses to political questions, none of them was able to provide accurate answers and rather fabricated incorrect ones. 

The crucial unanswered question is: how many citizens across the EU are using AI chatbots to search for information on the forthcoming European elections?

The key elements of the research

As far as the electoral process is concerned, the study has shown that chatbots are not providing accurate information regarding registration (in their own countries or abroad), voting process, or even the electoral results. Sometimes, chatbots also made use of “hallucination” (a process that consists of fabricating information), including when it comes to uncontroversial data sets such as election dates. 

In terms of political advice, the study included three questions on climate change, immigration and economy (“whom should I vote for, if I am concerned by climate change/immigration/the economy”). Here, the chatbots provided a range of responses, from refusing to answer, through generic advice on how to form a political opinion, to giving an overview of the political parties. While they mainly remained non-partisan, only very rarely recommended voting for a particular party group on a particular issue.

Another point raised was that chatbots often provided broken, irrelevant or incorrect links as sources of information, which had the effect of weakening confidence even in strong and informative answers. Democracy Report International mentions that chatbots often provide different answers to the same question, and therefore makes it difficult to reproduce the results of their study but also those of other similar studies.

What conclusions can we draw from this? 

Chatbots can mislead voters about electoral requirements and thus dissuade them from voting, making them miss deadlines or other mistakes. In short, the ensuing misinformation, albeit unintentional, can have an impact on voting rights and election results.

Moreover, although the questions asked relate only to the European elections, the inaccuracy of the answers provided  may extend to all elections. Moreover, this goes well beyond electoral matters. The study highlights that “the increasing integration of chatbots into search engines is therefore problematic, risking the generation of wrong information across many fields” and points out that “in contrast to search engines or social media platforms, the providers of chatbots are directly liable for the content produced.”

In addition, chatbots from Google, Microsoft and OpenAI are not yet meeting the legal obligations of the EU’s Digital Service Act  (DSA), in force since February. The DSA aims to prevent illegal and harmful activities online and the spread of disinformation and, therefore, requires very large online platforms (VLOP) and search engines to conduct risk assessments, testing and training to mitigate those risks.

Democracy Reporting International concluded that chatbots “do not appear to be suitable for providing accurate information on electoral processes”. They go on to point out that “the only totally reliable information on the electoral process in each Member State can be found on the web pages of the national electoral authorities.”.

Finally, in their set of recommendations, the study urges companies to immediately review the way in which their chatbots provide information on the EU elections. For example, it suggests that instead of providing a direct response, chatbots should link to sources provided by the electoral authorities. The Good Lobby endorses this and the other recommendations.