Earned Media Dominates AI Training: 84% of Citations Come From Journalism

Muck Rack's latest research reveals that traditional journalism and earned media overwhelmingly shape the information that generative AI systems cite and synthesize, according to the firm's third edition of its "What Is AI Reading?" study. The analysis of over 25 million links found that earned media drives a consistent 84% of all AI citations across the three leading large language models—ChatGPT, Claude, and Gemini—underscoring the continued centrality of professional journalism in the age of artificial intelligence.

The findings carry significant implications for public relations professionals, marketers, and companies seeking to influence how AI systems represent their industries and offerings. With paid and advertorial content accounting for merely 0.3% of citations, the research suggests that traditional earned media strategies remain the most effective path to shaping AI-generated narratives.

Study Methodology and Key Findings

Muck Rack's comprehensive analysis examined citation patterns across the three dominant AI platforms, revealing both broad consistency and striking platform-specific preferences in sourcing behavior.

The headline finding—that 84% of AI citations originate from earned media—held steady from previous editions of the study, indicating remarkable consistency in how AI systems prioritize different content types:

• Journalism accounts for 27% of all cited sources, making it the single largest category of references within AI training and response generation
• Paid and advertorial content represents just 0.3% of citations, suggesting these promotional channels have minimal influence on AI outputs
• Owned media and other sources comprise the remainder, with earned media's dominance spanning across all three major AI platforms

However, the research also uncovered significant differences in which specific sources each AI platform favors:

• ChatGPT cites Wikipedia most frequently, reflecting OpenAI's training data composition and the platform's preference for structured, encyclopedic information
• Claude shows a strong preference for PubMed Central, suggesting Anthropic's model has been trained with particular emphasis on scientific and medical literature
• Gemini frequently references Reddit, indicating Google's inclusion of user-generated discussion forums in its training datasets

These platform-specific patterns reveal that while earned media dominates overall, the specific sources within the earned media category vary considerably depending on each company's training data selection and model architecture.

Market Context and Industry Implications

The findings arrive at a critical moment when enterprises across sectors are grappling with how to position themselves in an AI-driven information landscape. The continued dominance of earned media—rather than paid promotional channels—represents a validation of traditional public relations strategies, even as companies invest heavily in AI and digital marketing.

The PR and communications industry has faced pressure to justify its value proposition as marketing budgets have shifted toward programmatic advertising, influencer partnerships, and owned social media channels. This research provides empirical evidence that quality journalism coverage remains the most effective mechanism for shaping how artificial intelligence represents organizations, products, and industries to end users.

The stark disparity between earned media's 84% share and paid content's 0.3% share reflects a fundamental truth about how AI systems are trained: they prioritize content that has undergone editorial review and journalistic gatekeeping. AI language models, trained on broad internet corpora and curated datasets, inherit the information hierarchies embedded in how human-created content is structured and distributed across the web.

This finding also has implications for the broader information ecosystem and content credibility landscape. As AI systems become primary research and information-gathering tools for millions of users, the sources these systems rely upon determine what information reaches the public. The continued reliance on journalism—despite its economic challenges and declining trust in some demographics—suggests that professional news organizations maintain outsized influence over AI-generated knowledge.

Investor Implications and Strategic Takeaways

For companies and investors monitoring the intersection of media, technology, and artificial intelligence, these findings carry several important implications:

Public Relations Services Face Reinvigorated Demand: The research validates the continued importance of earned media strategies, potentially supporting demand for PR agencies, media relations firms, and communications consultancies that specialize in securing quality journalism coverage. While these services have faced commoditization pressure, their ability to influence AI-generated narratives represents a new value proposition.

News Organizations Retain Strategic Importance: Publishers and journalism outlets benefit from their position as primary sources for AI training. As companies compete to shape AI outputs, demand for quality coverage in respected publications should remain robust, potentially supporting subscription models and B2B content licensing.

Paid Marketing Channels Face Headwinds: The minimal impact of advertorial and paid content on AI citations suggests that traditional advertising mechanisms—even when integrated into content—have limited influence on AI systems. This may prompt marketers to reconsider budget allocation and the relative effectiveness of promotional content at scale.

Platform Differences Create Strategic Opportunities: The variance in sourcing preferences across ChatGPT, Claude, and Gemini suggests that sophisticated communications teams might develop platform-specific strategies. Companies seeking to influence Claude outputs, for instance, might prioritize placement in peer-reviewed scientific journals, while those targeting ChatGPT users could focus on major encyclopedic resources.

These findings matter to a broad range of stakeholders:

• Technology investors monitoring how AI systems incorporate information and which corporate influences shape their outputs
• Media and publishing companies assessing their strategic position as AI training data sources
• Marketing and advertising platforms evaluating their effectiveness in an AI-mediated information landscape
• Public companies planning communications strategies to manage how AI systems describe their operations and offerings

Looking Ahead: The AI-Media Relationship

Muck Rack's study provides a crucial data point in the ongoing evolution of how information flows through AI systems and reaches end users. The consistency of the 84% earned media figure across multiple editions suggests these patterns may be relatively stable, at least in the near term. However, the distinct sourcing preferences across different AI platforms indicate that the relationship between artificial intelligence and information sources will likely continue to fragment and specialize.

As companies increasingly recognize that AI systems will mediate how their messages reach audiences, the strategic value of earned media coverage will likely remain elevated. The research suggests that the fundamental PR playbook—securing quality coverage from respected news organizations—retains its relevance and power, even in an era of AI-generated content and algorithmic information distribution. For organizations seeking to influence how artificial intelligence represents their industry, the path remains clear: invest in earning media coverage from credible journalistic sources.

The question for the coming years will be whether these patterns persist as AI training datasets evolve, as regulatory scrutiny increases, and as publishers develop new strategies for licensing their content to AI companies. For now, Muck Rack's research confirms that in the contest to shape AI narratives, traditional journalism remains the ultimate arbiter.