Earned Media vs. Owned Content: AI Citation Rates and Top Sources Ranked (2026)

Earned media is not 84 percent journalism. The 84 percent figure comes from Muck Rack's May 2026 What Is AI Reading? study of more than 25 million links cited by ChatGPT, Claude and Gemini, and it counts every non-owned, non-paid source: journalism plus encyclopedic, government, academic, social and other third-party content. Journalism alone is 27 percent. Muck Rack's own data and communications leads have written that they would not tell a communications leader the study means media relations drives 84 percent of AI visibility (Muck Rack in PR Daily, 2026).

The Machine Relations Index measures the same question by source role. In its October 17, 2026 release (mri_score_v2.1; 19,987 answer runs on six buyer question shapes across 19 categories, May 10 to October 17, 2026), editorial publications took 29.9 percent of the citations that went to a role-classified domain and company websites took 25.6 percent. Community platforms followed at 13.9 percent and academic and government sources at 12.0 percent; wire and press-release distribution took 0.8 percent. Those shares rest on the 28.4 percent of buyer-question citations that land on a classified domain. The rest went to a long tail of domains the Index has not assigned a role.

Concentration figures need the same care. Fifteen domains hold 68 percent of AI citations in the 5WPR AI Platform Citation Source Index 2026, pooled across every topic (5WPR/Everything-PR, May 2026). Measured within individual categories instead, the Index found the ten most-cited domains in a category hold between 7.0 and 24.2 percent of that category's citations (Machine Relations Index, mri_score_v2.0, September 23, 2026).

This article sets the Index's source-role measurement beside the third-party studies most often quoted on the question, and states what each one counts.

How the Machine Relations Index splits citations by source role #

The Index assigns each cited domain one of eight source roles where it has evidence for one, and leaves the rest unclassified. A company's own website is classed as vendor-owned whoever the company is, so for any single brand most vendor-owned citations are someone else's site: a competitor, a supplier or a platform. That is why a high non-owned share and a large vendor-owned share are not in conflict.

Source: Machine Relations Index release mri_score_v2.1+2026-10-17. Buyer shapes are best-of, how-to-choose, is-it-worth-it, problem-first, top-list and X-versus-Y questions in published strata. The whole release adds the legacy news-topic bucket, where company websites lead the classified roles at 33.7 percent against 19.3 percent for editorial publications.

Editorial publications lead the classified roles on every one of the six buyer shapes, from 24.7 percent on problem-first questions to 34.9 percent on best-of questions, but company websites are within three points on four of them. Across the 112 published buyer strata, editorial publications are the largest classified role in 44, company websites in 36, academic and government sources in 15, community platforms in 13 and market databases in 4. Among the 100 most-cited domains in the release, 26 are company websites, 24 are editorial publications and 2 are wire-distribution domains.

Two limits apply. The shares describe which sources were observed in answers, not what any placement produces. And from the October 15 release no domain is labelled an editorial publication on a name pattern alone, so a share in this release is not comparable with a September share without that change in view; 761 market-database labels still rest on a name pattern, and how much the labels move the rankings is measured separately.

Method note. Every figure in this section can be recomputed from the public release artifact at machinerelations.ai/data/machine-relations-index.json: the by_source_role block for the whole release, and each domain's per-stratum runs_cited for the buyer shapes. The release manifest at machinerelations.ai/data/mri-release-manifest.json carries the artifact hash.

The 15 domains that control 68% of AI citations #

The 5WPR Citation Source Index ranked the 50 domains most cited across ChatGPT, Claude, Gemini, Perplexity, and Google AI Overviews. The concentration is more extreme than organic search ever produced (5WPR/Everything-PR, May 2026).

Source: 5WPR AI Platform Citation Source Index 2026 (Everything-PR, May 2026)

The Similarweb January–February 2026 dataset of approximately 600,000 ChatGPT citation events in the U.S. provides a complementary single-platform view: Wikipedia (13.15%), Reddit (11.97%), OpenAI.com (6.21%), Walmart.com (2.90%), YouTube (2.67%), LinkedIn (2.42%), Reuters (2.27%), NIH (2.22%) (5WPR Citation Source Audit Q1 2026). Forbes is the only U.S. business publication in the top 20. The Wall Street Journal, Bloomberg, and the Financial Times do not appear.

Meltwater's GenAI Lens tracked over 8 million citations across eight LLMs in May 2026 and found the top five sources overall are YouTube (188,863 citations, up 56.4% month-over-month), Reddit (111,860), NIH (91,227, up 41.3%), Wikipedia (83,191, up 55.2%), and Statista (74,875) (Meltwater, June 2026). Forbes generated 55,975 citations, NerdWallet 40,752.

Which publications each AI engine cites most for consumer queries #

Each AI engine draws from a different source pool. The platform a consumer uses determines which sources shape the answer.

ChatGPT concentrates on institutional and encyclopedic sources. Its top-cited media outlets are Wikipedia, Axios, YouTube, Kiplinger, and Forbes (Muck Rack, May 2026). Axios appears in ChatGPT's top three cited domains across 13 of 17 industries — the only journalism outlet to achieve that consistency across any provider. Neither the New York Times nor Reuters appears in the top three for any industry. ChatGPT cites in 96% of responses with an average of 5 citations per answer. Earned and news media account for 51.1% of ChatGPT citations (Meltwater, June 2026).

Claude prefers long-form editorial outlets. Its top-cited media publications are U.S. News & World Report, Yahoo Finance, Bankrate, Consumer Reports, and The Motley Fool (Muck Rack, May 2026). The broader Claude editorial lean favors The New York Times, The Atlantic, The New Yorker, The Economist, Financial Times, and Washington Post (5WPR, May 2026). Only 36% of Claude's journalism citations come from the past 12 months, versus 56% for ChatGPT. Claude cites earned media in 65% of references and social content in 1% (University of Toronto, 2026). It cites in 55% of responses but averages 13 citations when it does.

Gemini draws from the most diversified source pool. Its top-cited media outlets are Forbes, Healthline, Business Insider, The Points Guy, and Substack (Muck Rack, May 2026). YouTube dominates Google's AI surfaces — it accounts for 29.5% of Google AI Overviews citations with a 200x advantage over other video sources (Everything-PR, June 2026). Roughly 43% of citations across Google's AI products point to Google-owned properties (YouTube, Maps, Scholar, Knowledge Graph). Reddit is Gemini's single most-cited non-Google domain at approximately 2.4% of all Gemini citations.

Perplexity favors primary research and community sources. Reddit dominates at 46.7% of top-10 citations, with Wikipedia at 19.8% (Profound, 2026). It also weights NIH/PubMed, Britannica, LinkedIn, and Quora heavily. Perplexity averages 21.87 citations per response and favors content published within the past 30 days.

Google AI Overviews split across YouTube (23.3%), Reddit (21%), and Wikipedia (18.4%). YouTube's non-ranking citation share is 18.8% — meaning it gets cited even when not in the organic results (AirOps, 2026).

Sources: Muck Rack May 2026, Meltwater GenAI Lens May 2026, 5WPR Citation Source Index 2026, Profound 2026

Cross-platform overlap is minimal. Only 11% of domains earn citations from both ChatGPT and Perplexity. Google AI Overviews and AI Mode share citations only 13.7% of the time. Between 78% and 85% of cited domains across the full dataset are unique to a single platform (Profound, 2026).

What the 84% earned media figure counts #

Muck Rack's What Is AI Reading? series has published three editions, in July 2025, December 2025 and May 2026. Across them, the broad pool of non-owned, non-paid sources ranged from 82 to 89 percent of cited links, and journalism stayed between 25 and 27 percent (Muck Rack, May 2026). Muck Rack labels the broad pool earned media and says it includes journalism along with encyclopedic sources, government sites, academic sources, social platforms and other third-party content (Muck Rack in PR Daily, 2026). Paid and advertorial content was 0.3 percent in the May edition.

Journalism citations span more than 20,000 distinct outlets with no single publication dominating — except Axios in ChatGPT. Industry trend queries drive journalism citations at more than double the rate of how-to queries. Press releases appear 3.5x more frequently in industry trend responses versus best-of queries.

An 85.5 percent figure also circulates, published by 5W and attributed to Muck Rack; the reviewed Muck Rack editions report 89, 82 and 84 percent, not 85.5 (source register). AirOps' October 2025 analysis of 21,311 brand mentions across 500+ top-of-funnel commercial queries found 85% came from third-party pages, and brands 6.5x more likely to be mentioned through third-party sources than through their own domains (AirOps, 2025). That is one vendor's sample.

Meltwater's GenAI Lens uses a narrower category: earned and news media accounted for 37.6% of all citations in May 2026 (down slightly from 39.5% in April), versus just 5.8% for social media and 0.2% for press releases (Meltwater, June 2026). While individual platforms like Reddit and YouTube rank as the top single-domain sources, the long tail of thousands of earned media outlets cumulatively represents the largest citation category.

Stacker reports a 239% median lift from its distribution network #

Stacker, which sells the distribution it measured, analyzed in a March 2026 study 87 stories across 30 brands, querying 2,600+ prompts across 8 AI platforms (Stacker, March 2026). The median lift from earned media distribution was 239%. Ninety-seven percent of distributed stories earned at least one AI citation, and 64% of all AI citations came from third-party publisher sources rather than the original brand domain.

The study expanded on Stacker and Scrunch's December 2025 pilot, which tested 8 articles across 944 prompt–platform combinations on five AI platforms and found a 325% lift (Stacker, December 2025). The broader March dataset confirmed the effect at lower magnitude but higher statistical confidence (p < 0.006).

The 19.2% syndicated-only figure from the December study remains the sharpest finding at the unit level: in nearly 1 in 5 answers, AI systems cited the third-party version and did not cite the original brand piece at all. The March 2026 study confirmed the pattern at scale — distributed versions were 5.3x more likely to be the sole source of AI visibility.

AI citations have almost no overlap with organic search rankings #

AI citations operate independently of organic search position. The gap has been measured by multiple independent studies and confirmed across platforms.

Moz analyzed nearly 40,000 search queries in February 2026 and found that 88% of Google AI Mode citations are not in the organic SERP for the same query (Moz, 2026). Only 12% of AI Mode citations match the exact URLs appearing in Google's top 10. Even at the domain level, only 1 in 5 citations come from websites that appear in the top 10 at all.

The University of Toronto's source-typology audit ran 1,516 queries across five systems (Google Search, GPT-4o, Claude 4.5 Sonnet, Perplexity Sonar Pro, Gemini 2.5 Flash) and found GPT-4o cites domains overlapping with Google's top-10 results at a mean of 4.0% (median 0.0%). All four AI engines showed under 16% mean overlap (University of Toronto, 2026).

Ahrefs found only 6.82% URL overlap between ChatGPT citations and Google's top 10 for the same queries (Ahrefs, September 2025). Profound's analysis of 650+ ChatGPT queries found 8–12% URL overlap depending on query type, with product queries showing a correlation of r ≈ –0.98 between ChatGPT citation frequency and Google rank (Profound, 2025).

BrightEdge's Generative Parser confirmed: only 17% of AI Overview citations come from pages that also rank in Google's organic top 10 (BrightEdge, February 2026).

Citation persistence at the URL level is 10.6% across 28 days — only 119 of 1,127 tracked URLs appeared in all three measurement waves. Maximum domain overlap between any two AI platforms: 17% (Perplexity / AI Overviews). Approximately 40–60% of cited sources rotate monthly (Profound, 2026).

Stacker reports distributed content keeps citations 2.1x longer #

Stacker's source decay research, analyzing over 3 million citation events across 120,000+ domains over a 26-week window (September 2025–March 2026), found that distributed content maintains citation authority for 2.1x longer than non-distributed content (Stacker, 2026).

The durability gap held across all eight verticals studied, ranging from +2.6 weeks in Tech & SaaS to +6.4 weeks in Real Estate.

What AI engines require to cite a brand #

Domain authority is the gateway. Ahrefs' analysis of ChatGPT's top 1,000 cited pages found 65.3% come from domains with a domain rating of 80 or higher, with a median DR of 90. Page-level link popularity barely matters (Ahrefs, September 2025).

BrightEdge found that ChatGPT includes brand mentions in 99.3% of eCommerce responses — but only for brands it has already resolved through sufficient independent credible coverage. A brand that exists only on its own domain is not a resolved entity and is absent from the citation layer regardless of domain authority or SERP position (BrightEdge, October 2025). Between 48% and 77% of citations across all AI engines come from specialized third-party sites — industry publications, niche experts, and topic-specific resources.

Data density amplifies citation rates in both earned and owned content #

The preference for earned sources operates independently from content structure, but content structure determines whether a given piece gets extracted when encountered.

The GEO paper (Aggarwal et al.) (Aggarwal et al., KDD 2024) tested 10,000 queries and found that adding statistics, quotations, and source citations each produced 30–40% improvement in AI visibility (Aggarwal et al., 2024). Keyword stuffing decreased visibility by 10%.

Meltwater's May 2026 data reinforces this: three of the ten most-cited journalists in their dataset come from Visual Capitalist, a publication focused on data-rich visual content. The pattern across all engines is consistent — data-rich, clearly sourced, visually structured content outperforms narrative-only formats in AI citation probability (Meltwater, June 2026).

Summary: the citation rate comparison in numbers #

The studies agree on one narrow point: answer engines draw most of their citations from sources outside the brand being asked about. They do not agree on how much of that is journalism. In the Index, editorial publications are the largest classified source role on buyer questions, and companies' own websites are close behind.

The Machine Relations Index republishes these source-role counts with every daily release, per category and question shape.

Frequently Asked Questions #

Which U.S. earned media sources are cited most by AI engines? #

The top earned media sources vary by platform. For ChatGPT, the leading journalism outlets are Axios (top 3 in 13/17 industries), Kiplinger, and Forbes. Claude favors U.S. News & World Report, Bankrate, and Consumer Reports. Gemini leads with Forbes, Healthline, and Business Insider. Across all engines combined, the 5WPR Citation Source Index ranks Forbes as the #5 overall domain and the top editorial source, followed by Business Insider, TechRadar, and Reuters.

How much does earned media distribution increase AI citation rates? #

Stacker, which sells distribution, reported a 239 percent median lift across 87 stories, 30 brands and 8 AI platforms in March 2026, and a 325 percent lift in an earlier 8-story pilot. Both are evaluations of Stacker's own network rather than independent measurements of earned media in general.

What percentage of AI citations come from earned media vs. brand-owned content? #

It depends on what is counted. Muck Rack's May 2026 study found 84 percent of cited links came from non-owned, non-paid sources, a category it calls earned media, with journalism at 27 percent. In the October 17, 2026 Machine Relations Index, editorial publications took 29.9 percent of role-classified citations on buyer questions and company websites took 25.6 percent. A company website in that count is usually not the brand being asked about.

Is there a dataset that classifies AI citations by source type? #

Yes. The Machine Relations Index assigns cited domains to eight source roles: editorial publication, vendor-owned source, community and social platform, academic and government source, market and company database, search or media platform, analyst and consulting research, and wire and press-release distribution. It publishes the counts per release, category and question shape in its public data file. Domains without enough evidence for a role stay unclassified, and those carried 71.6 percent of buyer-question citations in the October 17, 2026 release.

Do AI engines cite the same sources? #

No. Only 11% of domains earn citations from both ChatGPT and Perplexity. Between 78% and 85% of cited domains are unique to a single platform. YouTube dominates Google's AI products, Wikipedia dominates ChatGPT, Reddit dominates Perplexity, and Claude favors long-form editorial. Brands need multi-platform earned media strategies because each engine runs a different citation ecosystem.

Do AI citations last longer for distributed content? #

Yes. Stacker's source decay research found distributed content maintains citation authority for 2.1x longer — roughly 10 weeks versus 4.5 weeks for non-distributed sources. The persistence advantage held across all eight AI platforms and all eight verticals studied.

How do AI citation sources differ from Google search rankings? #

The University of Toronto found GPT-4o overlaps with Google's top-10 at a mean of just 4.0%. Moz found 88% of Google AI Mode citations are not in the organic SERP. Ahrefs found only 6.82% URL overlap between ChatGPT and Google's top 10. Ranking well in traditional search does not predict AI citation likelihood.