Earned media and AI visibility: which coverage gets cited?

In this article

When buyers ask ChatGPT for the best providers in a category, the answer often draws on external sources. Editorial coverage, analyst mentions, and trade publications can therefore shape how AI assistants describe a brand and whether they recommend it.

This guide covers what the available citation research actually measures, which coverage characteristics the data supports, and how to connect media work to a visibility measurement you can defend.

What the citation data shows

Muck Rack's What Is AI Reading? study analyzed more than 25 million links cited in ChatGPT, Claude, and Gemini responses across 17 industries. Its May 2026 edition reports that earned media accounts for 84% of cited sources, while paid and advertorial content accounts for 0.3%. Across three editions running from July 2025 to May 2026, the earned share stayed between 82% and 89%.

That headline number needs a careful reading before it goes into a management deck.

"Earned media" in this study means everything that is neither paid nor brand-owned. It includes Wikipedia, Reddit, PubMed Central, and community forums alongside news coverage. Journalism specifically accounts for 27% of cited sources, holding between 25% and 27% across the three editions.

The distinction matters because those two numbers support different decisions. 84% says that AI answers lean on third-party material rather than on brand-controlled pages. 27% is the share a media relations team is actually competing for. Quoting the first figure while planning against the second is how a real finding turns into a bad forecast.

Which coverage the data connects to citations

Four findings from the same study describe where journalism citations concentrate.

Read that last pair together. A spread across 20,000 outlets with almost no journalism title reaching a provider's top three suggests that broad, relevant, current coverage matters more than landing one flagship placement.

The study measures which sources were cited. It does not test why. The following placement characteristics are a reasonable inference from that data combined with what search providers document about retrieval, not findings the research establishes:

• coverage in publications that carry independent context for their field;
• current material, since AI systems favor sources they can treat as accurate now;
• specific, verifiable facts that a passage can support directly;
• consistent naming across outlets, so a brand resolves to one entity.

Google's own guidance is consistent with this. There is no special markup that guarantees inclusion, and meeting every technical requirement still does not guarantee that a page is crawled, indexed, or served.

Mentions and citations are different measurements

A mention means the brand appears in the answer text. A citation identifies a source the provider surfaced alongside the claim.

A brand can be recommended without any of its own pages being cited, and a page can be cited in an answer that never names the brand. Track the two separately, because they answer different questions and improving one does not automatically move the other.

Provider differences change what "visible" means

Citation behavior is not uniform, so a single-platform check misreads the picture.

The source preferences differ too. The study finds ChatGPT leaning on Wikipedia, Claude on PubMed Central, and Gemini on Reddit. A brand can look well cited in one product and absent in another without anything changing on its own website, which is why the provider belongs in the sample definition rather than in a footnote.

From coverage to a measurable connection

The sampling method is the one any AI visibility measurement needs, and What is AI visibility? sets it out in full: a fixed prompt set built from real stakeholder questions, several providers, repeated runs, written scoring rules, and a stable competitor set.

One step belongs to media work specifically. Once the answers are stored, match the cited URLs against your own coverage log. That comparison separates placements that reached readers from placements that are also feeding answers, and it is the only part of the workflow that connects a media plan to a citation.

Expect the match to be partial. Syndicated copies, aggregators, and paywalled articles complicate the mapping, and a cited page may carry a claim nobody on the team placed.

Report the provider, prompt set, market, language, date range, and competitor set alongside any percentage. The AMEC GEO Principles recommend a governed query library, repeated testing with variation disclosed, and a clear separation between visibility, outcomes, and impact. They also state plainly that no single score, tool, or prompt set proves total AI visibility.

Reading the connection honestly

Citation research describes a large sample of observed answers. It does not establish that a specific clipping caused a specific citation, and it cannot show that anyone read, believed, or acted on the answer they saw. Matching your coverage against cited URLs shows correspondence, which is useful for planning and weak as proof of cause.

The Barcelona Principles 4.0 require valid measures and separate reporting of outputs, outcomes, and impact. A citation count is an output.

Keep the claim proportionate to the evidence. Earned media plainly supplies most of the material behind AI answers, and journalism supplies a meaningful and stable share of it. That justifies measuring the connection. It does not justify attributing revenue to a citation count.

With aclipp, you can analyze clippings and AI visibility together. How AI answers find and cite sources explains the retrieval process, while What is AI visibility? covers the measurement model. For the broader PR context, see our guide to PR KPIs in 2026.