Over 95% of the cited links in AI responses come from unpaid sources, of which 85% have won the media. 27% is journalism
Miami, July 23, 2025 (Globe Newswire) – In a groundbreaking study analyzing millions of AI citation links from hundreds of thousands of prompts, Muck Rack, the leading provider of award-winning PR software, unveiled how AI prioritizes how it cites.
Starting to answer the basic research question entitled “What is AI Reading?”: Does media coverage have a significant impact on what AI is saying? The answer is clearly yes.
“This research is eye-opening for PR and communications teams, aiming to understand how media affects AI-generated content,” said Greg Galant, co-founder and CEO of Muck Rack. “We had theory and early signals up until now, and now there is solid evidence that media-acquisitioned media directly affects the output generated by AI.
Survey results
Citations actively affect AI output. Controlled prompt testing showed that LLM (large language models such as Openai’s ChatGPT and Google Gemini) changed significantly when citation was enabled. The media acquired not only appears, but it affects what is being said. This proves that the cited content is not decorative. Real-time dynamic inputs effectively ground the output.
The media acquired is basic input. Over 95% of the citations come from unpaid media sources, 85% of them come from sources that have been acquired, and another quarter comes from journalistic sources. Half of the AI total responses included at least one acquisition media quote. These numbers highlight the visibility of GEO (generating engine optimization) and AIO (artificial intelligence optimization) that has acquired media strategies.
What determines whether a brand is cited or not
Three important variables drive the inclusion of citations:
Remency: In particular, the Openai model prioritizes fresh content, particularly in topical, opinion-based or event-driven queries.
Query framing: Prompts based on advice or opinions trigger more dynamic quotes, but encyclopedia queries tend to rely on old static training data.
Outlet Authority: High-domain Orsolettes such as Reuters, Axios, Financial Time, AP, Time, Forbe, NPR, CNN are frequently cited, but among these, performance differs. Plus, more niche sites are frequently referred to, such as good housekeeping and Investopedia.
Strategic Impact on Communication Teams
Communication strategies need to evolve to explain this new layer of visibility.
New KPIs: Success includes traditional media placement and backlinking, as well as whether the brand is being cited in AI responses.
New Tactics: PR teams should prioritize highly authoritative outlets and tailor publishing cadences to the latest preferences of the AI model.
As AI increasingly mediates how information is accessed and consumed, it is clear that PR strategies must expand beyond human readers to include algorithm interpreters. The research confirms that many people in the industry are suspicious. The story you place, and where you place, is now taking shape on a large scale.
Galant added: “Generating AI is reconstructing itself at an unprecedented pace, and new patterns are constantly emerging, so thriving brands will be proactively tracking their positions and adapting in real time to stay ahead.”
To access the full report, go to GenerativePulse.ai/Whatisaireading.
Methodology
This study explores how modern generation AI systems cite sources in response to realistic user prompts. The aim was to quantify and characterize the nature of AI-generated citations across a variety of use cases and vendor models. This includes frequency, source type, and salience of acquired and owned media.
To achieve this, the generated pulses and McLac built a large, diverse set of prompts, ran them on several web-enabled language models, and then systematically analyzed the responses and citation links. The prompts span all industries and subjects. Sometimes they specifically mention businesses by name, but sometimes they don’t.
The following specific models were used in July 2025 to run hundreds of thousands of queries: chat GPT (both “4o” and “4o-mini”), gemini (‘flash’ and ‘pro’) and claude (‘sonnet’ and ‘haiku’).
###
About the generated pulse
Generic Pulse is a tool that helps PR and communications teams monitor and define how they appear in AI-generated search results. It builds around the concept of Generation Engine Optimization (GEO) and reveals which journalists, outlets and sources influence large-scale language models, such as ChatGPT. This allows teams to shape the brand’s visibility in the age of AI. Integrated into Muck Rack’s PR platform and backed by a $180 million Series A finance, Generative Pulse provides Comms experts with the leading insights and tools in AI-Native search situations. For more information, see GenericPulse.ai.
About McRuck
Muck Rack is a leading provider of award-winning PR software built around brands and agents. The only public relations software with intuitive technology and the most accurate and comprehensive data provided by journalists themselves, Muck Rack combines media databases, monitoring and reporting to combine seamless team collaboration, pitching and measurement. Dedicated for communication and public relations, Muck Rack helps over 5,000 companies around the world analyze and report the impact of media relations. Thousands of journalists use Muck Rack’s free tools to showcase your portfolio, analyze news on any topic, and measure the impact of your stories. For more information, please visit muckrack.com.
We’re hiring!
Contact: Communications Linda Zebian VP, Muck Rack linda@muckrack.com
