Earned media drives 84% of AI citations. That is not the argument for digital PR that it looks like.

The stat, and who publishes it

Muck Rack's May 2026 edition of "What Is AI Reading?" analysed more than 25 million links from ChatGPT, Claude and Gemini responses across 17 industries. It found earned media accounts for 84% of all citations. Paid and advertorial content accounts for 0.3%.

It is the third edition of an ongoing study, the sample is large, and the direction has held across all three. We have no reason to doubt the measurement.

What the 84% actually contains

The same study reports journalism separately, and journalism is 27%.

So roughly two thirds of the earned media figure is not journalism at all. The category as described spans academic research, government sources, encyclopedic sites and third-party corporate content alongside news.

That matters because digital PR, as sold, targets journalism. It does not target PubMed. It cannot place you in a government dataset. It does not write your Wikipedia entry, and attempting to is against Wikipedia's rules.

The honest one-line summary is that paid and advertorial content is a very small share of what these systems cite, at 0.3%. Small is not zero, and the study does not establish that the remaining 99.7% is independent editorial endorsement. Earned, unpaid and independent are three different properties, and this figure separates paid from unpaid rather than sorting the rest.

Earned coverage and an AI citation are different objects

The phrase "earned citations" gets used as though it were one thing. It is two, and they can occur without each other.

An answer can cite your own site, an earned article, a paid placement, or a syndicated copy of any of them. Record the displayed URL and classify the relationship separately, because the same citation count can mean very different things depending on what is behind it.

The top domain per engine

This is the part that should change how the stat is used. The most cited domain, per model:

An encyclopedia, a biomedical research archive, and a forum. Not one of the three is a digital PR target. You cannot pitch Wikipedia, you cannot place a client in PubMed Central, and pitching Reddit in the PR sense is a good way to be removed from it.

Within journalism, the concentration is severe too. Citations span more than 20,000 outlets, but Axios alone appears in ChatGPT's top three cited domains across 13 of 17 industries.

So the realistic addressable slice for a business without an existing press profile is a fraction of the 27%, competing against outlets that dominate by default.

The gap between finding and pitch

Every step here is defensible except the last one.

• AI systems cite earned sources far more than paid ones. Measured, large sample, consistent across three editions.
• Therefore visibility in AI answers depends on being present in sources models draw from. Reasonable.
• Therefore buy digital PR. This does not follow.

The third step assumes the sources models cite are the sources PR can reach. The study's own breakdown says they mostly are not.

A measurement of where information comes from is not a measurement of what you can change.

The strongest positive finding, and who published it

The case against is not the whole picture, and the best evidence for distribution comes from a company selling it.

Stacker published research on 16 March 2026 covering 87 stories across 30 brands, queried on 8 AI platforms with roughly 30 prompts per story. It measured citations to the brand's own domain and to publisher sources in its network, before and after distribution, and reported a median lift of 239% in AI visibility, with cross-platform coverage moving from 5.4% to 17.9% at the median.

What it demonstrates is narrower than it first reads. Getting more copies of your story onto more domains puts more citable surfaces in front of retrieval, and those surfaces get cited. That is a genuine mechanism, and it is a different claim from independent editorial endorsement.

Syndication breadth and third-party corroboration are not the same asset. A hundred copies of your own press release is one source repeated. A trade publication that examined your work and wrote about it is a second opinion. Both can produce a citation. Only one of them tells a reader anything they could not have got from you.

The confounder nobody controls for

Brand mentions correlate with AI visibility. The obvious alternative explanation is fame, and we have not seen it controlled for.

Well-known companies get written about more often. Well-known companies also get recommended more often. A correlation between mentions and recommendations is exactly what you would expect if both are consequences of being known, with no causal path from one to the other.

That does not mean mentions do nothing. It means the studies published so far cannot separate the two, and the direction of the arrow matters enormously if you are deciding whether to spend a budget on outreach.

The critical survey of GEO research published in July 2026 reviewed 45 studies and concluded that no technique demonstrates stable cross-platform effects on organic discovery. Digital PR is not exempt from that finding.

What is actually addressable

The study still points somewhere useful, just not where it is usually pointed. Read the source types as a list of places to be correct and present, rather than places to pitch:

• Encyclopedic and reference presence. Not writing your own Wikipedia entry, which will be reverted. Making sure the factual record about your business is consistent and verifiable wherever it already appears.
• Community and forum presence. Gemini's top domain is Reddit. Being genuinely useful in the places your buyers ask questions is slower than a press release and considerably more durable.
• Directories and review platforms. Third-party corporate content sits inside the 84%. A complete profile with real reviews is earned media in the sense that matters here, and it is achievable this month.
• Trade and niche press over national. You will not displace Axios. A specialist outlet your buyers read is a realistic target and a more relevant citation.
• Being the primary source. The most reliable way into earned media is to publish something worth citing. That is slower than outreach and it does not need anyone's permission.

If your question is narrower, namely whether your own coverage is the thing being cited, that is an attribution problem with its own answer. For most small businesses the honest ordering is: fix the factual record, get real reviews on real platforms, then consider outreach. Press coverage is the last of those, not the first, and the study most often used to justify going first does not support it.

The correlational evidence behind "mentions beat backlinks" has the same sampling caveat worth reading, and whichever way you go, measure it against a dated baseline rather than a feeling.

Limits

• We read Muck Rack's published summary of the study, not the underlying dataset. Figures are quoted as published there in May 2026.
• The summary we read does not give a formal verbatim definition of the earned media category, and states no methodological limitations. Both are worth noting when a number is repeated this widely.
• Citation share is not traffic and it is not revenue. A cited domain is not a customer.
• These systems change. A May 2026 measurement describes May 2026.
• We have not run a controlled test of outreach against a holdout. Nobody we can find has published one either, which is the actual state of the evidence.

If someone publishes a controlled study showing outreach moves citation rates for businesses without an existing press profile, that would change the recommendation here and we would say so.