- Muck Rack's "Generative Pulse" study (three editions since July 2025, 25M+ cited links analyzed across ChatGPT, Claude, and Gemini) has found earned media accounts for 82-89% of all AI citations in every edition. Paid and advertorial content: 0.3%.
- AirOps' 2026 analysis of 21,000+ brands found brands are roughly 6.5x more likely to be cited via a third-party source than their own domain, with about 85% of brand mentions in AI answers sourced from third-party pages.
- Within that third-party share, listicles, comparisons, and review platforms dominate: roughly 90% of third-party citations, per AirOps' data. A separate 30,000-citation dataset (BuzzStream/Citation Labs) found press releases specifically earn only about 0.04% of AI citations, corroborating Muck Rack's near-zero paid-content figure from an entirely independent dataset.
- A fifth independent study, 5W's "State of AI Search 2026" (10,000+ buyer-intent prompts across five platforms), found third-party validation outperformed both paid media and owned content in every category it tested, not just on average.
- What's unconfirmed: why this happens. The explanation offered across GEO industry analysis (that independent sources carry authority/corroboration signals owned content can't structurally replicate) is a reasonable theory, not a disclosed mechanism from any AI lab. This piece is explicit about that line.
Ask most AI-visibility vendors what to invest in first and the pitch defaults to content: publish more, publish better, publish faster. The actual data on where AI citations come from doesn't support that as the primary lever. Across every independent, disclosed-methodology study that's measured it, earned media (content a brand didn't write and doesn't control) dominates citation share by a wide, consistent margin.
This is Stage 4 of the five-stage sequence, and it's the stage with the most consistent evidence behind it of any in the framework. What follows is the full body of that evidence: not just the headline number, but what corroborates it, what doesn't hold up, and what's still unknown about why it's true.
What's the core finding, and does it replicate?
Muck Rack's Generative Pulse team has published three editions of its "What Is AI Reading?" study since July 2025, each analyzing tens of millions of AI-cited links across ChatGPT, Claude, and Gemini. Earned media (journalism, third-party reviews, academic and government sources) has accounted for 82-89% of all citations across those editions. Paid and advertorial content has stayed near-zero throughout: 0.3%. Journalism alone has held consistently at 25-27% of all cited links across every edition.
That range holding steady across three separate measurement windows, using the same disclosed methodology each time, is a meaningfully stronger form of evidence than a single snapshot study: the same measurement, repeated, landing in the same place.
Do four more independent studies agree?
If Muck Rack's numbers stood alone, they'd be worth real weight already. They don't stand alone: several structurally different studies, from different organizations, using different sampling methods, point the same way:
AirOps' 2026 State of AI Search report analyzed 21,000+ brands and found brands are roughly 6.5x more likely to be cited via a third-party source than their own domain, with about 85% of brand mentions in AI answers sourced from third-party pages. Within that third-party share specifically, AirOps found roughly 90% of citations come from listicles, comparisons, and review sites, and 80% of cited brands appear in the first three positions of whatever list or comparison cited them.
5W's "State of AI Search 2026" ran 10,000+ buyer-intent prompts across ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews, spanning 15+ categories with quarterly year-over-year tracking. Its finding: independent third-party validation from high-authority publications outperformed both paid media and owned content as a citation input in every category tested, not an average result skewed by one strong vertical.
BuzzStream/Citation Labs' dataset (30,000 citations across 595 prompts and 10 industries) found press releases specifically earn only about 0.04% of AI citations, an independent dataset landing almost exactly where Muck Rack's 0.3% paid-content figure sits. Two unrelated organizations, two different datasets, the same near-zero result for paid/self-published promotional content.
One caveat from the same dataset: BuzzStream found blog/content format was the single largest citation category overall at roughly 53%. That's a different cut than earned-versus-owned: "blog" here is a content-format label, not a distinction between a brand's own blog and someone else's, so it doesn't contradict the earned-media finding above, but it shouldn't be read as an owned-content counter-finding either. The two numbers are answering different questions.
What doesn't this mean?
This isn't an argument that owned content doesn't matter. It's a precondition, not a competing lever: content that isn't structured to be extractable doesn't get cited regardless of who's pointing at it, and entity resolution has to hold before any citation, earned or owned, attaches to the right node in the first place. What the data argues against is treating owned-content volume as the primary lever for citation growth, when four independent measurements agree the majority of citation share is earned, not published by the brand itself.
What's still open: why?
Every study above measures the same thing: earned media outperforms owned content by a wide margin, consistently, across platforms and time. None of them explain the mechanism, and that distinction matters, because the explanation that circulates in GEO industry writing (that independent, third-party sources carry corroboration and authority signals a brand's own content structurally can't self-supply, since a brand describing itself isn't independent evidence of anything) is analyst reasoning, not a disclosed algorithmic fact from OpenAI, Google, Anthropic, or Perplexity. It's a plausible, widely-repeated explanation, not a confirmed one. Treat the finding, earned beats owned, consistently, by a wide margin, as solid. Treat any specific explanation of why, including this one, as a reasonable theory still waiting on a lab to confirm or refute it directly.
Frequently asked questions
Does this mean I should stop investing in my own content?
No. Content extractability is a precondition, not a competing strategy. Poorly-structured owned content doesn't get cited regardless of what else you do, and your own site is still where entity signals (schema, sameAs, canonical facts) live. The finding here is about where the majority of citation volume actually originates once the foundation is solid, not a case for abandoning owned content.
What counts as "earned media" exactly?
Coverage or mentions from a third party that you didn't pay for and don't control the framing of, journalism, independent reviews, third-party comparison and "best of" content, forum and community discussion. Paid placements and advertorial content are a separate, near-zero category (0.3% of citations per Muck Rack's data) despite sometimes being marketed as functionally similar.
Do review platforms like G2 or Capterra count as earned media?
Yes, and they're a large share of it. AirOps' 2026 analysis of third-party citations found roughly 90% come from listicles, comparisons, and review sites specifically. A review isn't something you write yourself; it's a third party's independent account, which is exactly the profile of content that performs best in this data.
Why do AI models cite third-party sources over brand-owned content?
The honest answer is that this isn't fully settled. The finding itself, that earned media dominates citation share, is well-established across multiple disclosed-methodology studies. The explanation for why (that independent sources carry authority signals brand-owned content structurally can't self-supply) is industry analysis, not a confirmed mechanism disclosed by any AI lab. Treat the what as solid and the why as a reasonable, unconfirmed theory.
Sources referenced
- Muck Rack Generative Pulse, "What Is AI Reading?" (Jul 2025 / Dec 2025 / May 2026 editions, 25M+ links)
- AirOps 2026 State of AI Search report
- 5W State of AI Search 2026
- G2 review-burst citation-lift study (Kevin Indig) (Jun 2026, 2,514 products)
This piece is the full expansion of Stage 4 in the five stages for winning in the AI search era. For the mechanics of what makes owned content citable in the first place, see how to write content that AI models can actually quote; for the entity-resolution precondition, see what is entity consistency. For a specific mechanism this off-site work likely feeds, a brand's odds of making a model's pre-search shortlist, see is your brand on the AI buyer's shortlist before anyone even searches?
