What AI Is Actually Reading. And Why It Matters More Than Your Last Press Release.
I have a confession. I was a Star Trek fan club member before The Next Generation aired.
I have been to conventions. I attended a Klingon wedding, which is exactly as intense as it sounds and significantly louder than a human one. And I have worked with George Takei, which remains one of the more surreal and genuinely wonderful experiences of my professional life. The man is as warm and hilarious in person as you hope he will be.
This is my life. Live long and prosper.
I mention all of this not to establish my nerd credentials, though they are considerable, but because after reading Muck Rack’s new Generative Pulse study on how AI platforms cite sources, one phrase kept coming to mind: Space, the final frontier.
Except the frontier in question is not the cosmos. It is the increasingly strange and consequential territory of what one might call brand accuracy in AI. By that I mean what AI is actually reading about your brand, and whether what it finds resembles the truth.
A new study just answered a question that has been quietly haunting every PR professional for the past two years: When someone asks an AI chatbot about your industry, your company or your competitors, what does AI actually reach for?
Spoiler: It is not your carefully crafted press release. It is also not your award-winning thought leadership content. And it is definitely not that blog post your CEO wrote in 2019 that took six weeks and four rounds of legal review to publish.
The data has opinions, and they are not gentle ones:
- Across more than 25 million analyzed links from ChatGPT, Claude and Gemini, earned media dominates. About 99% of AI citations come from non-paid sources, with journalism alone accounting for 27% of all references.
- Paid and advertorial content? A rounding error of 0.3%.
- Press releases have fallen to just 1.1% of citations, down sharply from 6% as recently as December 2025.
The headline takeaway for communications professionals is not subtle: If you are not generating earned media, you are largely invisible to AI. All that money you spent on sponsored content is doing a lot of heavy lifting for your ego and essentially nothing for your AI footprint.
Three Platforms, Three Completely Different Information Worlds
ChatGPT cites sources in 96% of its responses, averaging five sources per response. Gemini cites sources in 82% of responses, averaging eight sources for each response. Claude is the most selective, appearing in only 55% of responses, but when it does cite, it averages 13 sources per response.
These are not stylistic differences. They represent fundamentally different epistemologies about how to answer a question. Think of them as three starships running the same mission with completely different crew directives. They are not all reading the same newspaper over breakfast.
The source divergence is just as striking:
- ChatGPT’s top cited domain is Wikipedia.
- Claude’s is PubMed Central.
- Gemini’s is Reddit. Yes, Reddit. The same platform where strangers debate whether a hot dog is a sandwich is now shaping what one of the world’s leading AI systems tells your customers about your industry. Mr. Spock would find this deeply illogical.
Axios cracks ChatGPT’s top three cited domains across 13 of 17 industries studied. The New York Times and Reuters do not appear in the top three for any industry. If you have been spending your media relations budget chasing the Times, we would like a word.
For clients in the tech sector, the data shows trade publications dominate over mainstream news for both ChatGPT and Claude. This aligns exactly with what we have observed in our own platform analysis: AI models are drawing from a narrower, more specialized media diet than most clients assume. Think trade press, not front page. Think niche, not national.
The Question Type Changes Everything
Perhaps the most actionable finding in the report is this: Industry trend queries drive journalism citations at more than twice the rate of how-to queries.
When someone asks what is happening in an industry, AI reaches for news coverage. When they ask how to do something, AI leans on owned content and reference material instead.
This distinction has direct implications for how campaigns are structured. Press releases, despite their diminished role, still appear disproportionately in industry trend responses, at rates more than 3.5 times higher than in best-of queries. Timing matters. Topic framing matters. The question your audience is likely asking matters. Structure your campaign around how people actually query AI, not around what looks good in your monthly report.
What RealSense Taught Us (The Hard Way)
At Bospar, we arrived at most of these conclusions not from a study, but from a crisis. Which, as crises go, was at least an instructive one.
When we helped spin RealSense out of Intel, we secured more than 500 stories and drove a 4x increase in website traffic. By traditional PR metrics, it was a success worth celebrating. Then we scanned the AI platforms with something closer to a tricorder than a media monitoring report, and what we found was alarming: ChatGPT described the company as dead.
Outdated Intel links and broken references were shaping the narrative. Our client, very much alive and actively selling technology, was being eulogized by a chatbot. Red alert.
That crisis led to the development of Audit*E, our proprietary GEO platform, which monitors how AI platforms represent a brand, scores their accuracy and identifies the specific source material driving the problem. The Muck Rack study validates the framework we built to solve it.
You cannot influence what AI says about you without first understanding what it is reading.
If your company is asking:
- Why is my company invisible to AI search?
- Why does AI say wrong things about my company?
- How do press releases perform in AI search?
- What content does AI cite most?
- How does AI decide what to say about a brand?
Audit*E and this new study can provide some of the answers.
GEO Is Not Optional
The study’s finding that more than half of journalism citations come from articles published within the past 12 months should recalibrate how clients think about the long-term value of any given placement. Recency is a factor. Volume is a factor. But as the Axios anomaly demonstrates, brevity, cross-sector coverage and consistent publishing cadence matter too.
Axios appears in ChatGPT’s top three domains for 13 of 17 sectors because of how it is written, not just due to audience size. Short, clear and everywhere. There’s a lesson in there somewhere.
What Audit*E confirms, and what the Muck Rack data reinforces, is that AI visibility and AI brand representation are not byproducts of good PR. They are a discipline in their own right.
The clients who treat GEO as a measurement layer on top of their existing media program will continue to be surprised by what the chatbots say. Sometimes unpleasantly.
Resistance is futile. The AI is reading and directing buyers whether you are ready or not.
