Five AI Assistants Show Vastly Different Approaches to Political Recommendations
ChatGPT and Grok recommend political parties when asked how to vote. Copilot, Gemini, and Perplexity refuse to do so more than 90% of the time. A study of 1,220 queries across four languages reveals that AI assistants are becoming political intermediaries-with inconsistent guardrails and measurable real-world implications for how people encounter political information.
Researchers from the Universitat Oberta de Catalunya and Science4Insights submitted the same 61 questions to five AI platforms between January 9-11, 2026. Each question focused on voter concerns documented by Spain's Centre for Sociological Research: housing, jobs, healthcare, immigration, and unemployment. The researchers asked in Spanish, Catalan, Galician, and Basque, then tracked which parties each platform recommended.
Who Gets Recommended
ChatGPT made a political recommendation in 43.2% of queries. Grok did so in 35.8%. The other three platforms stayed neutral more than 90% of the time.
When recommendations were made, the Socialist Party (PSOE) came out consistently ahead across all platforms and languages. Sumar and Podemos trailed far behind. The People's Party (PP) appeared frequently in results-mentioned 96.3% of the time-but received direct recommendations only 28.2% of the time, suggesting visibility doesn't equal endorsement.
Vox had high visibility but low recommendation rates, always below 6%.
Language Changes the Answer
The language of the query shaped which parties appeared. Catalan queries returned an average of 4.78 parties per response versus 3.82 for Spanish. Regional parties like ERC and Junts appeared far more often when users asked in Catalan. EH Bildu and the PNV dominated Basque-language results. BNG had greater presence in Galician.
This matters: a voter in Catalonia asking in their native language receives different information than one asking in Spanish-from the same AI platform.
What Sources Drive Recommendations
The assistants cited Wikipedia most frequently (958 mentions), followed by El País (620) and Spain's state broadcaster RTVE (505). Party websites also appeared regularly. Gemini cited external sources less than 1% of the time, while ChatGPT and Grok accounted for the vast majority of citations.
This concentration matters. A party's visibility in a handful of reference sources directly affects how AI presents it.
What This Means for Communications Professionals
The researchers offer four takeaways for PR and political communications teams:
- Monitor AI assistants as part of active listening. They're now part of how voters encounter information.
- In multilingual markets, assume different audiences get different answers. A single communication strategy won't work across language groups.
- Visibility in AI results doesn't equal endorsement. Being mentioned frequently doesn't guarantee favorable positioning.
- Each AI platform behaves differently. Strategies that work for one won't work for all.
The study shows that AI for PR & Communications professionals now includes managing how conversational assistants present political information. Unlike search engines, where rankings are transparent, AI recommendation patterns remain opaque to most users. That opacity matters when voters are making decisions.
The research is currently undergoing publication review.
Your membership also unlocks:
