What is AI Reading? Generative Pulse by Muck Rack

What content do AI systems pick up when they generate answers and what criteria do they use to select their sources? The report What is AI Reading? by Muck Rack investigates this question. The analysis is based on more than one million links evaluated from responses from ChatGPT, Claude, Gemini and Perplexity, collected between July and December 2025.

The results show that journalistic content and earned media play a central role. Around 94% of the links cited by AI come from unpaid reporting, with journalistic sources accounting for around a quarter. Earned media dominates above all for open, explorative questions, while owned media is predominantly used for concrete, fact-oriented queries. At the same time, it is clear that AI models respond to topicality. More than half of all citations relate to content from the past twelve months, with the highest citation probability seven days after publication.

The growing importance of press releases is also highlighted. Their share of AI citations has increased fivefold since July 2025. The report shows that cited press releases are often clearly structured, contain more robust statistics, use bullet points and make greater use of action words.

Another key factor is source authority. AI models take into account both the reputation of large leading media outlets and the professional authority of specialized industry sources. Platforms such as Reddit are also frequently used, but usually in combination with journalistic content.

The results make it clear that the citation practices of generative AI are constantly changing. Between summer and winter 2025, all of the models examined adapted their source preferences. For example, content from management consultancies became less important, while visual platforms such as YouTube became more relevant, especially for Gemini. Effective GEO therefore requires ongoing adjustments to topicality, authority and the structural design of content.