The public debate about AI and journalism has so far focused heavily on copyright, licensing, newsroom automation and the use of generative AI in content production. These issues are important. But another transformation is already taking place: AI systems are becoming intermediaries between citizens and the news.

When a user asks an AI system what is happening in a country, which outlets are reliable, who owns the main television channels, or whether public media are independent, the answer may shape how that user understands the information environment.

The system may cite a narrow group of dominant outlets, rely on international sources instead of local journalism, ignore investigative or minority-language media, amplify state-aligned sources, or provide confident answers without transparent attribution.

For media pluralism, this creates a new research challenge. It is no longer enough to ask who owns media outlets, who funds journalism, or which platforms distribute news. We also need to ask which sources AI systems make visible and which ones they leave out.

The AI Pluralism Monitor is designed as a growing country database. The first phase focuses on ten countries selected to capture variation in language, media-system structure, market size, platform dependence, political environment and media-capture risk.

The database is then expanded over time to include additional countries, regions and language environments. This allows the project to track not only how AI systems treat individual national media ecosystems, but also how patterns of source visibility and information diversity compare across countries.

The first group of countries combines European and non-European cases, large and medium-sized media markets, English and non-English environments, EU and non-EU countries, and contexts with different levels of media freedom, public-service media strength and political pressure.

For each country, MJRC builds a verified baseline list of relevant media sources, including national news outlets, public-service and state media, television and radio brands, digital-native outlets, local and regional media, investigative and non-profit outlets, minority-language or community sources, and politically aligned or state-influenced media where relevant.

Researchers then use a standard prompt bank to ask each AI system comparable questions in English and in the main local language. The prompts cover current news, reliable source recommendations, media ownership, public media, local and minority-language media, political influence, corruption, elections and media freedom.

Each response is archived with the full prompt, system name, date and time, language, links, citations, screenshots where needed, and information on whether browsing or search mode was active.

The project then codes every response for the sources cited, mentioned or recommended; the accuracy of the information provided; the transparency of attribution; the type of sources used; and the diversity of the media ecosystem represented.

The project does not start from the assumption that AI systems are biased against particular outlets, countries or political positions. Instead, it uses structured testing to document patterns of visibility, omission, concentration, distortion and attribution.

The project distinguishes between sources that are not cited, sources that are misrepresented, and sources that are systematically made more visible than others.

It also separates factual accuracy from pluralism, because an AI answer can be factually correct while still relying on a narrow or unrepresentative source base.

The project publishes its methodology, prompt bank and coding framework so that its findings can be scrutinised, replicated and expanded.