With AI changing discoverability at an unprecedented rate, we’ve been connecting the dots to see what’s trending, what’s changed, and what’s staying the same.
In July 2025, Muck Rack’s seminal AI report, Generative Pulse, transformed hunches about how LLMs work into usable data, tracking how AI is shaping discoverability and brand authority from data beginning in December 2024. They’ve continued to release updates as the state of AI changes; the latest release dropped in May 2026.
Stacker has been taking notes on these changes, too, through our original Citation Lift research, expanded study, and more which examine how rates of brand discoverability are affected in LLMs when brands amplify their brand journalism through wider distribution.
More than two years after AI became a constant refrain in business, and about a year from the moment it entered the mainstream workforce, we take a look back at how Muck Rack’s findings on AI have evolved and the implications for brands investing in Stacker’s earned distribution model.
AI Trends We’ve Observed – (What Hasn’t Changed, Much)
Since the first major report by Muck Rack that analyzed how AI models processed and surfaced data, one of the biggest throughlines is:
Earned media coverage is still a huge influence on AI responses
Earned media’s pull on GEO & AI answers
When GEO became mainstream, research showed that many of the sources cited by major LLMs came from earned media, though not all. The rate of LLMs sourcing earned media content citations has fluctuated a bit over time, but has generally remained above 80%. So, while it may not be all it certainly is an overwhelming majority.
*Note that the Muck Rack report defines “earned media” as journalism, along with academic research, government and NGO sources, encyclopedic sites like Wikipedia, social platforms, and third-party corporate content. Though journalism remains the biggest piece of this Earned Media Pie, currently contributing 27% of earned media citations.
More recent reports show that citations carry more weight than sourcing, but can actually “fundamentally alter AI responses,” according to Muck Rack. This illustrates how important earned media and earned distribution are in shaping the way users view, understand, and consume information. Stacker’s own research builds on this, showing that it’s not just whether you’re cited, but how widely your content is distributed across credible outlets that determines how often and where you appear.
This is what we’ve termed coverage breadth, which has emerged as a key driver of visibility across LLMs and GEO KPI.
“They found that enabling citations doesn’t just add sources, it changes the response itself.” — Muck Rack
💡Keep in mind: The way users phrase their queries impacts the results they get, so brands should carefully consider the types of queries they prioritize and make sure their published content reflects those queries. The May 2026 report found that when users ask questions about industry trends, journalism citations appear more than twice as often as in “how-to” queries. Competitive citations about industry trends are usually published within the past six months.
That’s also why distribution matters: broader editorial coverage increases the number of entry points your brand has across different query types and prompts.
What’s Changed in the Data
Though we know earned media and distribution are must-haves for brand visibility and authority, the inner workings of LLMs are constantly changing.
Muck Rack acknowledged in a February blog that changes in LLM algorithms are changing so rapidly that their data releases should be treated as “time capsules.”
Models use citations with varying depth and frequency
The most recent Muck Rack report highlights behavioral differences between AI models: ChatGPT includes citations in 96% of its responses — with about 5 sources per response — whereas Claude relies more on training knowledge. Claude only includes citations in 55% of its responses, but when it does, it uses 13 sources on average. Gemini averages eight sources when it cites, 82% of the time.
Top-cited publications
What publications each AI model cites has also further fragmented, according to the May 2026 Muck Rack report.
The May 2026 release showed very little overlap in citations generated by journalism among the top AI models, except for the media outlet Axios, which was cited in ChatGPT, Claude, and Gemini across 13 of the 17 industries Muck Rack is tracking.
See the latest report for in-depth breakdown into which journalists and media outlets are preferred by the top AI models for each industry.
💡Hot tip: Reading between the lines of these GEO research reports, brands are wise to let these broader LLM trends nudge their content strategies, rather than doing a 180.
Getting one top-tier outlet doesn’t mean success across all systems, either, so prioritize diversified media strategies and be sure to monitor discoverability for each LLM.
Definitions of recency
In March 2026, new Muck Rack insights showed major differences in how the leading AI platforms define recency. The definition of recency has accelerated slightly between March and May 2026, the report says.
LLMs still favor recently published content — but the definition of timeliness has changed.
The first Generative Pulse report found that the major AI platforms, Claude and ChatGPT, preferred to surface sources under a week old. Now, Claude prefers sources from around 10 weeks ago, while ChatGPT still prefers content published less than 1 week ago. This aligns with newer research on citation persistence: distributed editorial coverage doesn’t just increase initial visibility, it extends how long content remains citable as models refresh their answers.
The report found that journalism is the key source for identifying recent, credible information.
💡Keep in mind: These findings explain why your brand might not be showing up in certain models. The data also reinforces the need for a consistent drumbeat of content that can appeal to both platforms — an added bonus is that earned media discoverability compounds over time, especially when it's distributed.
What’s Clear: Earned Media is Shaping Answers & Citations
Despite the speed of change, brands working to shape their authority in LLMs understand that prioritizing earned media strategies above all is the wisest investment they can make in 2026.
Breadth of media coverage is key
A single top-tier media hit doesn’t have staying power in GEO.
What was once a big win is now a flash in the pan and will get buried within days if the coverage isn’t amplified. Our source decay analysis quantifies this shift, showing that content supported by broader editorial distribution maintains visibility in AI systems far longer than one-off placements.
Coverage breadth = new authority signal
Learn more about coverage breadth
Stacker provides brands with a content distribution platform built for earned distribution to enhance their coverage, with over 3,000 participating news outlets using Stacker to fill their coverage gaps.
Above all, regular, high-quality editorial coverage and reach show LLMs that your brand’s content is worthy of learning from and reward it by pushing it to the top.
The emerging pattern across both external research and Stacker’s own studies is clear: visibility in AI search is no longer about a single moment of coverage, but about sustained, distributed presence across trusted sources.
Want to see how your content can reach new heights like Paylocity?
Let’s chat about how Stacker’s distribution platform can get you the discoverability and staying power you need to succeed and show up in AI search results.
Doriane is Stacker's Director of Marketing. She previously headed marketing departments in various companies, led a marketing agency for 7 years, and built a startup combining automated marketing workflows with a marketing services marketplace.
Photo Illustration by Stacker // Canva
