The New Gatekeepers: What AI’s Source Selection Means for Communications

At a Glance
_A new study by the German think tank Agora explores how leading AI platforms select sources when answering news-related questions. The findings suggest that AI does not draw on a broad cross-section of the media landscape. Instead, each system consistently relies on a relatively small set of preferred publishers, with established news organizations serving as the primary source of information.
As AI increasingly becomes a gateway to news and information, understanding how these systems select and prioritize sources is becoming a strategic issue for communicators._

How AI Chooses Its News Sources

Large language models such as ChatGPT, Claude, Gemini, Google AI Overview and Perplexity regularly cite sources when responding to questions about current events. But where do these references actually come from?
Agora set out to answer this question by prompting the five systems with requests for current news on domestic politics, foreign affairs, business, climate and migration. Typical prompts included questions such as “What are today’s latest developments on [topic]? Please include links to your sources.”
The study analyzed 4,811 citations generated across 675 prompts covering politically and socially relevant topics.

A Small Number of Publishers Receive Most of the Attention

Although the researchers identified 554 different domains across all responses, only a small group of publishers accounted for the vast majority of citations.
Most sources appeared only occasionally—and many only once. In other words, AI systems follow a classic power-law distribution: a handful of highly visible publishers dominate, while hundreds of other sources remain largely invisible.
Rather than broadening the diversity of information, AI appears to reinforce existing patterns of media attention.

Public-Service Media Hold a Strong Position

One of the study’s most notable findings is the prominent role of Germany’s public-service broadcasters.
Tagesschau.de is the most frequently cited source in both Google AI Overview and Perplexity, while Claude most often references ZDFheute.de and Gemini favors Deutschlandfunk.de.
Their strong editorial reputation undoubtedly contributes to this visibility. Equally important, however, is the fact that their content is freely accessible rather than hidden behind paywalls, making it easier for AI systems to reference.

ChatGPT Is an Outlier

Among the models analyzed, ChatGPT stands apart.
Its responses overwhelmingly cite Welt.de, which, according to the study, appears more frequently than the other nine sources in the top ten ranking combined. Overall, Welt.de received 382 citations, making it the single most referenced publisher in the entire dataset.
The authors suggest that OpenAI’s partnership with Axel Springer may be one factor influencing this outcome.

Google AI Overview Delivers the Broadest Range of Sources

Google AI Overview generated the highest number of source links overall, with 1,261 citations across all prompts. It was followed by Perplexity (930), Claude (900), ChatGPT (777) and Gemini (around 700).
However, the total number of citations does not necessarily indicate source diversity, since the same domains are often referenced repeatedly.

Looking instead at unique domains paints a different picture. Google AI Overview cited 258 different websites, Claude 209, Gemini 173, ChatGPT 120 and Perplexity just 117.
The conclusion is consistent across all systems: they repeatedly rely on a relatively limited pool of trusted sources.

Journalism Remains AI’s Primary Knowledge Base

Across every model examined, journalism remains the dominant source category.
ChatGPT and Perplexity rely particularly heavily on editorial media, with 77% and 73% of their citations respectively coming from journalistic outlets. Gemini, Claude and Google AI Overview reference government institutions more frequently, including Germany’s Federal Government, Parliament and Foreign Office websites.
For broad news queries, traditional journalism continues to be the backbone of AI-generated answers.

What This Means for Communications Professionals

For communicators, the implications are significant.
AI systems largely build their answers on existing media coverage. If an organization, brand or executive is absent from credible editorial reporting, they are far less likely to appear in AI-generated responses.
This elevates the strategic importance of media relations. In an AI-driven information environment, earned media is no longer just about influencing public perception—it increasingly shapes what AI systems “know” and subsequently surface to users.
The concentration of citations among a small group of influential publishers further amplifies this effect. Visibility in leading media outlets now extends beyond traditional audiences and directly affects discoverability within AI-powered search and answer engines.
It is worth noting that Agora’s research focused on broad news topics typically covered by national and business media. In more specialized industries, the picture is likely to differ. Trade publications, owned media and authoritative third-party platforms may play a much greater role in shaping AI responses.

AI Source Selection Remains a Black Box

Despite their growing influence, AI providers remain largely opaque about how they select, rank and prioritize sources.
The information users receive depends not only on the AI platform itself, but also on the topic and even the precise wording of a prompt.
The study also highlights another challenge: AI systems do not exclusively cite reliable journalism. Claude, for example, repeatedly referenced content from the Russian disinformation network Pravda without distinguishing it from legitimate news organizations.

The Bottom Line

There is no AI shortcut to visibility.
Organizations looking to strengthen their presence in AI-generated answers should continue to invest in high-quality journalism, credible media relationships and authoritative content.
As AI becomes an increasingly important gateway to information, strong editorial visibility is emerging as one of the most effective ways to remain visible—not only to audiences, but also to the AI systems that increasingly shape public discourse.