vote_box_cropped_1200x1200.jpg

The AI Dimension of Political Reputation

November 19, 2026

Europe is entering a consequential electoral cycle. Beginning with Spain at the end of November 2026 and continuing with France, Italy, Greece and Poland in 2027, national elections will help shape the future direction of the European Union.

They will unfold against an unusually volatile geopolitical, economic and information environment.

From a geopolitical perspective, unresolved international crises continue to fuel political polarisation, as shown by the divisions over whether to extend support for Ukraine and increase defence spending, and by the political tensions surrounding the situation in Gaza.

From an economic perspective, voters are increasingly mobilised by widespread inflation, the quintessential “government consensus killer,” as well as by the housing crisis and growing social inequality.

From an information perspective, setting aside the pressing issue of potential foreign interference that may amount to forms of cognitive warfare, the most significant development is the rapid adoption of generative artificial intelligence.

Every day, people turn to AI chatbots not only for answers, but increasingly to access information, understand complex issues and form opinions. This creates a new challenge for governments, political parties and political leaders: understanding how AI platforms interpret and represent their positions to voters.

Political reputation, in other words, increasingly has an AI dimension.

This challenge was the starting point for a study conducted by APCO using its proprietary RepGenAI methodology, focusing on Italian political debate ahead of the 2027 general election.

The study tested four major AI platforms, ChatGPT, Microsoft Copilot, Google Gemini and Google AI Overview, asking a series of questions repeatedly over a one-month period. The prompts did not directly mention political parties, focusing instead on three issues likely to feature prominently in the electoral campaign: the economy, foreign policy and immigration.

More than 8,000 responses generated over 30,000 spontaneous mentions of political parties, allowing us to examine two distinct dimensions: visibility and recommendation.

On visibility, the four platforms present a relatively consistent picture. Five parties dominate: Fratelli d’Italia/Brothers of Italy (FdI), Lega/League, Forza Italia, Partito Democratico/Democratic Party (PD) and the Movimento 5 Stelle/Five Star Movement (M5S), each totalling between 39% and 48% of mentions. Azione (13.7%), the Greens and Left Alliance/AVS (13.4%) and Italia Viva (9.2%) follow further behind.

This convergence largely disappears when AI platforms move from party mentions to recommendations. When the same question was put to all four platforms, in 67.6% of cases at least one prioritised a different party from the others.

Being visible in AI-generated answers, therefore, does not necessarily mean being recommended. Different models can reach different conclusions because they assess and combine information using different sources and criteria.

The picture also changes by issue:

  • On the economy, including jobs, wages and taxation, parties belonging to Italy’s progressive alliance (the so-called “Campo Largo”) collectively received 47.2% of recommendations, compared with 41.6% for the governing centre-right alliance.
  • On foreign policy, the centre-right led, with FdI receiving 25.5% of recommendations, followed by Forza Italia at 17.3% and the League at 14.1%.
  • On immigration, the advantage was clearer still: FdI and the League together accounted for 48% of recommendations.

Importantly, these recommendations were often conditional. In 47.7% of cases, platforms linked their answer to the user’s priorities, effectively suggesting: if this issue matters most to you, this is the party closest to your position.

One of the study’s most significant findings concerns the sources behind these answers. Each platform appears to draw from a distinct information ecosystem: no pair shared more than 23.7% of the sources cited in their responses.

Interestingly, those sources include political parties’ own websites. Once seen as relatively traditional digital communication tools, party websites may acquire renewed importance as sources for generative AI. ChatGPT was particularly notable: 92% of all citations to political-party websites captured by the study came from its responses.

That raises a new question for political communicators. Content created for human audiences increasingly has a second audience: the AI systems that retrieve, interpret and synthesise it.

The picture provided by the study is necessarily a snapshot and may evolve over time. New political forces show why: because they have a more limited information history, they might appear less visible in AI answers than opinion polls might suggest. One example is the far-right party Futuro Nazionale, which accounts for just 2% of mentions in the study. Qualitatively, however, those mentions already reveal a strong thematic profile: immigration and national sovereignty together account for almost half (47.6%) of them.

But a broader implication is already clear. Political communications has traditionally focused on what politicians say, what journalists report, what search engines surface and what social media algorithms distribute. Generative AI adds another layer: what AI systems synthesise from those sources and return to voters as an answer.

Political stakeholders will therefore increasingly need to understand how their policies and reputations are represented across AI platforms, what information ecosystems shape those representations and how they can effectively disseminate their messages.

For voters, the challenge is equally important. AI literacy will increasingly mean understanding not only whether an answer is accurate, but why different platforms may provide different answers to the same political question.

Understanding that new information environment may soon become an essential part of understanding electoral campaigns themselves.