A recent study conducted by researchers from the Open University of Catalonia (UOC) in collaboration with Science4Insights has revealed the growing role of AI assistants as information intermediaries in democratic processes, specifically in influencing voters’ decisions.
The research, which analyzed over 1,220 queries in multiple languages (Spanish, Catalan, Galician, and Basque), showed that some platforms, such as ChatGPT and Grok, tend to offer specific electoral recommendations rather than remain neutral, posing significant challenges to public relations and political communication strategies.
These findings corroborate previous studies from 2025, which indicated that brief conversations with AI trained on specific political arguments can indeed shift voters’ convictions and voting intentions.
Is AI a new political intermediary? The study, titled “Breaking the Algorithm?”, involved posing 61 key questions to five major platforms: ChatGPT, Copilot, Gemini, Grok, and Perplexity. The questions focused on issues affecting ordinary Spanish citizens, such as housing, unemployment, immigration, and healthcare, based on data from the Center for Sociological Research (CIS). This data was collected between January 9 and 11, 2026, to test the algorithms’ responsiveness to an explicit request to recommend a specific political party.
Disparity in Neutrality: ChatGPT vs. Grok
The results revealed a sharp divergence in the platforms’ behavior. Looking at the most biased, ChatGPT topped the list, providing electoral recommendations in 43.2% of cases, followed by Grok at 35.8%.
As for the most neutral platforms, Gemini, Copilot, and Perplexity maintained near-complete neutrality, refusing to provide recommendations in over 90% of the queries.
What does AI base its decisions on?
Intelligent search engines rely on specific reference sources, with Wikipedia leading the way with 958 citations, followed by major national media outlets such as El País and RTVE. Interestingly, Gemini cited the fewest external sources (less than 1%), while Grok and ChatGPT relied entirely on external references to construct their responses.
Recommendations for Political Communication Experts
Based on these findings, the researchers advised professionals in the political sector to actively monitor what AI assistants say as part of their listening strategy.
Regarding multilingualism, it is important to recognize that audiences will receive different responses depending on the language they use when speaking to the machine.
