Yes. Muck Rack's third What Is AI Reading? study, analyzed by Known & Cited's GEO Research 2026, found that 84% of citations across more than 25 million links in ChatGPT, Claude, and Gemini came from earned media. That figure has stayed between 82% and 89% since July 2025.
What "Earned Media" Actually Means in This Data
Before drawing conclusions, the definition matters. Muck Rack counts journalism, academic research, government sources, encyclopedias, and third-party company content as earned. Journalism alone accounts for 25-27% of citations, a figure that has held steady across three editions of the study. Paid and advertorial content sits at 0.3%.
A competing measurement from Meltwater, which analyzed 5.35 million citations in April 2026 and sorted domains differently, reportedly put earned and news media at 39.5%. Both figures can reflect reality depending on how categories are drawn. The directional conclusion holds either way: brand-owned pages and paid placements are a small fraction of what AI models actually cite.
Query Stage Changes Everything
The 84% headline is an aggregate. Dig into query type and the picture shifts sharply.
At the B2B evaluation stage, Ten Speed's September 2026 research found 88% of citations pointed to content brands control. BrightEdge found brand-owned pages took 42-79% of consideration-stage citations across eight industries. Peec AI research reported by Search Engine Journal in June 2026 found that the format of a question alone shifted brand visibility by 20-25%.
This is the detail most summaries skip. Earned media dominates broad and top-of-funnel queries. Owned content recovers ground at the comparison and decision stage. An AEO strategy that treats all queries the same will misallocate effort.
Why Owned Pages Struggle at the Top of the Funnel
The mechanism is how answer engines retrieve and filter sources. Perplexity, for example, runs a retrieval-augmented generation loop: it searches the live web, ranks candidates by relevance, freshness, and source trust, and then passes the strongest results to a language model that writes the answer. Sources treated as trustworthy tend to be mainstream news outlets, reference sites like Wikipedia, and community forums.
Self-referential brand pages don't disappear from AI answers, but they face a structural disadvantage on informational queries where corroborated, third-party coverage carries more weight. A brand page that says "we are the best option" competes against a trade publication that reached the same conclusion independently. The AI has little reason to prefer the brand's own claim.
Ahrefs' May 2026 experiment reinforces this. Adding JSON-LD schema to 1,885 pages produced citation changes of +2.2% in ChatGPT and +2.4% in AI Mode, both statistically indistinguishable from zero. Schema is not the lever many practitioners hoped it would be.
The Volatility Problem
One further complication: AI citation lists are unstable. SparkToro and Gumshoe.ai had 600 volunteers run 12 prompts through ChatGPT, Claude, and Google's AI nearly 3,000 times. The Search Engine Land summary of that study found the probability of getting the same list of brands twice was under 1 in 100. The same list in the same order was closer to 1 in 1,000.
What was stable: how often a brand appeared across many runs for a given intent. Frequency of citation across repeated queries is a more reliable signal than any single answer. That has direct implications for how you measure GEO performance.
Where Budget Should Go
Given the data, the case for redirecting resources is straightforward.
Digital PR, expert quotes in trade press, and third-party review sites are the channels most likely to produce the citations that AI answers pull from. Profound's September 2026 research found that only 1 in 4 recommended brands appear in answers across different user personas, down from 2 in 5 for the same persona. Breadth of third-party coverage across different editorial contexts increases the chances of appearing across more persona-specific queries.
On-site schema optimization is not worth significant ongoing spend given the Ahrefs findings. The technical basics are worth doing once, correctly. After that, the work that moves AI citation rates is mainly off-site.
Monitoring Tools and What to Look For
The Adweek roundup of best-funded GEO startups reflects how much capital is now chasing the problem of measuring AI visibility. Several tools in the AEO space have developed earned-media citation tracking as a specific feature.
Tools like Profound, Peec AI, OmniSEO, and Evertune AI track citation share and mention frequency across AI platforms. Athena HQ and PromptWatch offer share-of-voice and visibility monitoring across repeated query runs. When evaluating any of these, the relevant question is whether the platform measures citation frequency across multiple runs of the same query rather than reporting on single-answer snapshots, which the SparkToro/Gumshoe data suggests are unreliable on their own.
Caveats Worth Keeping
The 84% figure comes from one vendor's dataset with one definition of earned media. Orbit Media's September 2026 analysis found only 1.7% of question-and-website pairs where ChatGPT, Claude, Gemini, and Perplexity all cited the same site for the same question. That cross-model disagreement matters: a citation strategy built around one AI platform may not transfer to another.
The practical implication for AEO practitioners: earned media coverage in credible third-party outlets is the most durable investment across the current AI citation landscape, but the stage of the buyer journey and the specific AI platform still determine which source types actually appear in the answer. Neither owned nor earned content dominates everywhere. Measure both, separately, across query stages.