Earned Media Beats Backlinks for AI Citations, New Correlation Data Shows

Every link-building retainer in this industry was built around the same premise: more backlinks, higher domain rating, better rankings. That premise still holds for classic Google rankings. It does not hold for AI citation, and now there's a number attached to exactly how much it doesn't hold.

The number that should worry every link-building budget

Jaxon Parrott's Machine Relations Research citation factors study, published July 4, 2026, ran Spearman correlations between a set of off-site brand signals and AI citation and visibility. The underlying data pulled from an Ahrefs analysis of 75,000 brands with Domain Rating over 40, filtered to keywords with 800 or more monthly search volume. That's not a small sample, and it's not cherry-picked for a narrative. It's a broad look at what actually moves the needle when an AI engine decides which brand to cite.

The result: backlinks, the asset most link-building programs exist to acquire, correlated with AI citation at just 0.218. YouTube mentions correlated at 0.737, the strongest signal in the entire dataset. Branded web mentions came in at 0.664. Branded anchor text landed at 0.527. Brand search volume sat in the 0.334-0.392 range. Backlinks were dead last, and not by a small margin. They correlated at roughly one-third the strength of branded mentions.

Read that table again with a specific question in mind: if you had to defend your current link-building budget line by line to a CFO using only this data, which row would you point to? Almost nobody's budget is weighted toward YouTube mentions or branded web presence right now. Almost everyone's is weighted toward backlinks, the row at the bottom.

What actually correlates with AI citation

The pattern across the top three rows isn't subtle. YouTube mentions, branded web mentions, and branded anchor text all outrank raw backlinks, and none of them require a link to exist at all. A mention of your brand name on a YouTube channel, in a forum thread, or in an article that never links to your site still correlates more strongly with getting cited than a standard backlink does. That's the opposite of how most digital PR and outreach programs are scoped today, where the deliverable is almost always defined as "a link," and a mention without one is treated as a missed placement rather than a win.

This lines up with what we've already documented in our breakdown of the six link types AI engines actually trust: the links AI models lean on hardest aren't the ones link builders have historically chased. Directory submissions and guest-post placements, the bread and butter of a lot of link-building retainers, sit far from the top of that trust hierarchy. This new correlation data adds a second, independent confirmation from a different research methodology.

Parrott's research also found content depth matters on its own axis: content with 19 or more data points earns two to three times more AI citations than text-only content. That's a separate lever from any of the off-site signals in the table above, but it's worth naming here because it means the fix isn't purely an off-site play. Dense, data-rich content earns citations on its own merits, and it's also the kind of content that tends to get mentioned and shared, which feeds the branded-mention signal from a different angle.

Earned media still owns the citation pie

Muck Rack's third-edition "What Is AI Reading?" study, last updated May 7, 2026 and authored by VP Communications Linda Zebian, analyzed more than 25 million links across ChatGPT, Claude, and Gemini responses spanning 17 industries. The headline finding: earned media accounts for 84% of all AI citations. Journalism alone, a subset of that earned-media share, accounted for 27% of cited sources on its own. Paid and advertorial content, by contrast, showed up in just 0.3% of citations, a rounding error next to everything else in the dataset.

Muck Rack cofounder and CEO Greg Galant put it plainly: "For communications teams, that is validation that the work they do to earn coverage in the right outlets has real consequences." That's not spin. It's the same story the correlation data tells from a different angle. Whether you're measuring raw citation share (Muck Rack) or signal correlation (Machine Relations Research), earned coverage and branded presence outperform anything paid or purely link-transactional.

The two studies also converge on the same range independently. Machine Relations Research's own supporting citation notes that earned media accounts for 82-89% of all AI citations, a band that sits right on top of Muck Rack's 84% figure despite coming from different underlying datasets. That kind of agreement between two separately run studies is rare enough to take seriously.

Muck Rack's data also breaks down citation behavior by engine, and the differences matter for how you prioritize coverage. ChatGPT cites sources in 96% of responses, averaging 5 citations per response. Gemini cites in 82% of responses, averaging 8. Claude cites in only 55% of responses, but when it does, it averages 13 citations, the deepest sourcing of the three. If your buyers are living in Claude conversations, getting cited at all matters more than in ChatGPT, where citation is nearly universal but shallower per response. This citation-behavior split is part of why domain concentration in AI answers looks different depending on which engine you're auditing.

One more number from the supporting research is worth flagging on its own: only 38% of AI Overview citations now come from Google's organic top 10 results. That's a separate finding from the correlation and citation-share numbers above, but it reinforces the same conclusion. Ranking well in classic search and getting cited by AI are increasingly different games, played with different inputs.

Why backlinks still matter, just not for this

None of this means backlinks are worthless. That would be an overcorrection, and this data doesn't support it. Backlinks remain a real input to classic Google ranking signals, and Domain Rating still tracks with organic visibility in traditional search results. That's a separate question from AI citation, and conflating the two is exactly the mistake this data corrects.

The distinction that matters is which job you're hiring a link-building budget to do. If the job is ranking on Google's organic results page, backlinks still belong in the mix, though even there the Semrush figure on organic-top-10 overlap suggests the returns are shrinking. If the job is showing up as a cited source inside ChatGPT, Claude, or Gemini responses, a correlation of 0.218 is a hard number to justify spending the bulk of a budget against, when 0.664 and 0.737 sit right there in the same dataset.

This is the exact gap our guide to link building for the AI era was written to close: the tactics that built domain authority for the last decade aren't the tactics that earn AI citation now, and running one program to chase both goals with the same backlink-count KPI means under-serving whichever goal isn't Google rankings.

The decision framework: where the budget goes now

Strip this down to what a budget owner actually needs to decide. Three signal categories, three different jobs: raw backlink acquisition (guest posts, directory links, generic outreach for a link) is the weakest correlate of AI citation in this dataset at 0.218. Earned media and journalism coverage is where the majority of citation share already sits, at 84% per Muck Rack, with Galant's read being that genuine editorial work "has real consequences." Branded off-site presence, specifically YouTube mentions at 0.737 and branded web mentions at 0.664, correlates more strongly with citation than anything link-shaped in the dataset, whether or not a link is attached.

The verdict: if backlink count or Domain Rating is still the primary KPI on your link-building program, that KPI is optimizing for the weakest of the three levers, for AI citation purposes specifically. It should move to secondary status for GEO work, not disappear, secondary. The primary budget lines should be earned coverage (press, journalism, the kind of placements our link-building service is built to land) and branded off-site presence, with YouTube treated as a first-class channel rather than an afterthought bolted onto a video-marketing budget somewhere else in the org chart.

This reprioritization matters more, not less, for categories where the buying decision leans on trust signals outside a vendor's own website. Tech and SaaS brands are a clean example: buyers in that vertical already treat third-party validation, reviews, comparison threads, analyst mentions, as part of the evaluation process, which means the branded-mention and earned-media channels this data points to are channels those buyers are already reading before they read your site.

Practically, this means a link-building retainer built for 2026 GEO work should report on mentions and coverage volume alongside link counts, not instead of them, and it should treat a strong branded mention on a high-traffic YouTube channel as equal or greater value to a standard backlink, even when no link is attached. If your current reporting has no line item for branded mentions or YouTube presence at all, that's the gap this data says to close first.

Do this next

Pull your last two quarters of link-building deliverables and sort them into the three buckets above: raw backlinks, earned media placements, and branded mentions (including YouTube). If the raw-backlink bucket is more than half the total, that's the budget line to renegotiate this quarter, not next year. Redirect that share toward earned coverage and branded presence, and start tracking mentions as a KPI in its own right, separate from link count, so the next quarterly review actually reflects what's driving AI citation instead of what used to.

See where you are cited today

A free snapshot audit of your rankings and AI citations before we ever talk.