Earned Media Drives 84% of AI Citations — 3 Editions Above 80%
Muck Rack's third What Is AI Reading? report says 84% of 25 million+ cited links across ChatGPT, Claude, and Gemini came from earned media, the sources brands neither own nor pay for — 82% to 89% in all three editions. Jaxon Parrott on why repeated source-composition evidence changes the operating model.
Earned media, sources brands neither own nor pay for, supplied 84% of the cited links in Muck Rack's May 2026 study across ChatGPT, Claude, and Gemini; journalism alone was about 27%. That share has held at 82–89% across three editions since July 2025. Paid and advertorial content together account for 0.3%. Muck Rack's May 2026 What Is AI Reading? report, based on more than 25 million links across 17 industries, confirmed it for the third time.
Three editions make this an operating signal.
When a metric holds across three measurement windows, it is an operating signal. AI engines do not publish a setting that says "prefer earned media." Muck Rack's repeated finding shows that sources brands neither own nor pay for dominate what AI engines cite, so teams should measure whether independently published evidence is retrievable for their own categories.
Muck Rack's CEO Greg Galant put it directly: "Three editions in, the data keeps telling the same story: earned media is what AI trusts." Source: https://muckrack.com/blog/what-is-ai-reading-may-2026
The per-engine breakdown tells you where the pressure concentrates:
| Engine | Cites in % of responses | Avg citations per response | Top cited domain |
|---|---|---|---|
| ChatGPT | 96% | 5 | Wikipedia |
| Gemini | 82% | 8 | |
| Claude | 55% | 13 | PubMed Central |
Claude cited sources in 55% of the sampled responses and averaged 13 sources when it did cite. Those are observed frequency and depth measures; they do not establish that Claude has the "highest bar" or that a brand survives a hidden provider filter.
One data point stood out across the study: Axios appeared in ChatGPT's top three cited domains in 13 of 17 industries.
What the paid and owned categories show.
Paid and advertorial content was 0.3% of the cited links in Muck Rack's May 2026 study, press releases 1.1%, and first-party corporate, blog and owned content 13.7%.
Ahrefs' analysis of 75,000 brands found that branded web mentions correlated 0.664 with Google AI Overview visibility, while backlinks correlated 0.218 in the same study. The measured unit is cross-brand correlation in that dataset; it does not establish that mentions cause AI visibility or that third-party context is the only relevant source role. Source: https://ahrefs.com/blog/ai-overview-brand-correlation/
That difference is useful diagnostically: teams should measure brand mentions, backlinks, cited source roles, and answer visibility separately rather than treating any one correlation as the provider's causal rule.
Owned content remains useful for canonical facts, product detail, and conversion. The operating question is whether independent, community, reference, and owned sources together give answer systems enough attributable evidence for the target query.
Google expanded five citation and link surfaces.
On May 6, Google announced five updates to how AI Mode and AI Overviews surface links. Inline links, hover previews, subscription labels, "Explore new angles" sections, and community perspectives create more visible routes to a mix of publisher, community, reference, and owned sources. The announcement does not say every surface rewards earned media. Source: https://blog.google/products-and-platforms/products/search/tools-partnerships-web-ecosystem
The practical implication is to inspect which source roles actually appear in those surfaces for the category, rather than assuming Google excludes brand-owned or paid pages by design.
Syndication is citation infrastructure now.
Stacker's partnership with Scrunch tested syndication across identical articles and reported up to a 325% citation-rate lift in the first-look pilot, from 8% to 34%, across eight stories and 944 prompt-platform combinations on five tested platforms. A later 87-story analysis reported a 239% median lift. The measured unit is citation-rate change inside the tested distribution cohorts. This does not establish a universal syndication multiple, a private provider mechanism, or a guaranteed citation, recommendation, pipeline, or revenue outcome. Source: https://stacker.com/blog/how-earned-media-distribution-expands-ai-visibility-first-look-at-citation-lift
More than half of journalism citations in the Generative Pulse sample came from articles published within the prior 12 months, with fewer observed after six months. That is a recency distribution in the sample, not a universal decay law; teams should measure their own categories over time before setting refresh or coverage cadence.
This is what I built AuthorityTech to do.
AuthorityTech's operating thesis is that independently published evidence can give answer systems more attributable material than unsupported self-assertion alone. The studies above make that thesis testable through source-role measurement; they do not prove one universal trust mechanism.
That is why I coined Machine Relations in 2024: to measure and improve how owned, earned, community, and reference sources make a brand retrievable, attributable, and citable when discovery is mediated by machines.
AuthorityTech uses a pay-per-placement model: the fee sits in escrow, and it is refunded in full if the agreed placement is not delivered. That model depends on the relationships to deliver — 1,500+ direct editorial contacts built over eight years. No cold outreach. No pitch queue. One call.
FAQ
What is the Muck Rack Generative Pulse report?
The Generative Pulse is Muck Rack's ongoing study of cited-source composition. The May 2026 edition analyzed over 25 million cited links across ChatGPT, Claude, and Gemini, covering 17 industries. Across editions, Muck Rack reported 82–89% of cited links in its broad earned-media category of non-owned, non-paid sources, which includes journalism alongside reference, institutional, community, UGC, and other third-party sources. Source: https://muckrack.com/blog/what-is-ai-reading-may-2026
Who coined Machine Relations?
Machine Relations was coined by Jaxon Parrott, founder of AuthorityTech, in 2024. It names the discipline of measuring and improving how brands are retrieved, described, cited, and recommended through owned, earned, community, and reference sources.
What does the report show about paid sources?
Paid and advertorial content was 0.3% of cited links in Muck Rack's May 2026 study, and about 99% of cited links came from non-paid sources.
How do different AI engines handle citations differently?
In Muck Rack's sample, ChatGPT cited in 96% of responses and averaged 5 sources, with Wikipedia as the top domain. Gemini cited in 82% and averaged 8, with Reddit as the top domain. Claude cited in 55% and averaged 13 when it cited, with PubMed Central as the top domain. These are observed sample differences, not disclosed authority or recency thresholds.
What should brands do about this data?
Audit which engines cite you and your competitors, classify the cited source roles, and test whether stronger owned evidence, independent coverage, community discussion, or reference sources change the answer set. Treat the studies as baselines for measurement, not guarantees that one placement type drives AI visibility. Run a free audit at https://app.authoritytech.io/visibility-audit to see where you stand.