Does PR still matter in the AI era?

What does AI actually cite when it answers a buying question?

The most useful place to start is not a theory about PR but a count of what these systems quote.

Muck Rack’s What Is AI Reading? study analyzed more than 25 million links cited by ChatGPT, Claude and Gemini across seventeen industries. Earned media — journalism, expert commentary, third-party write-ups — accounted for about 84% of all citations. Journalism alone was roughly 27%. Paid and advertorial content came to 0.3%.

What makes the figure worth taking seriously is its stability. Across three editions of the study going back to July 2025, earned media has stayed in a band of 82–89%, and journalism between 25–27%. A number that holds through model updates and a full year tends to describe how these systems work, not a passing quirk.

So the question PR keeps being asked — whether it still matters once buyers ask an AI instead of searching — has an empirical answer worth sitting with. When the engine writes the answer, it is mostly reading the coverage, or noticing its absence.

Why would an engine trust the press over a brand’s own site?

Because a model is reluctant to repeat a claim it can find in only one place — the place that benefits from it.

A company’s own pages are where it states its facts. Earned coverage is where those facts get confirmed by a party with no stake in the sale. That second step — independent confirmation — is what an engine seems to weigh before it puts a brand into an answer.

There is supporting evidence beyond the citation counts. The Princeton group that first named generative engine optimization found that content backed by citable sources and real statistics earned roughly 41% more AI citations than the same content without them. The reward, in other words, goes to claims a machine can verify against something outside the brand.

This is less a new law than an old instinct, automated. A careful reader has always trusted the independent write-up over the brochure. The answer engines have simply made that instinct the default, at scale.

If earned media wins, do press releases still do anything?

The wire blast on its own barely registers. The genuine coverage it can earn is the entire point.

The uncomfortable figure here comes from Meltwater, whose April 2026 analysis of roughly 5.35 million citations across eight major engines put press releases at about 0.4% of everything cited — a rounding error next to earned and news media, which held near 39.5% in the same dataset. Distributing a release across a newswire, by itself, does almost nothing for AI visibility.

That is not an argument against the press release; it is an argument about where its value sits. What an engine cites is rarely the release itself but the work it can set in motion — a reported article, an analyst’s note, a credible third-party piece that states a clear fact in plain text. The job did not disappear. It moved from distribution toward earning the kind of corroboration a machine will quote.

What does this change about how PR should be measured?

The instinct to count clippings sits awkwardly against how engines actually choose whom to name.

If the goal is to be cited, then the more honest gauge is whether the coverage is the kind an engine can lift: a specific, checkable claim, stated plainly, in an outlet the model already reads. A glowing feature that never states a clear fact about the brand is harder to quote than a short, factual mention in a trade title.

Two habits seem to follow from the data, and both are modest. The first is to pitch for a quotable fact rather than for column inches — coverage that says something an engine can extract. The second is to keep the same claim consistent everywhere it appears, so a model is not left adjudicating between three versions of the same company. Consistency, here, reads as confidence.

None of this asks PR to become a different discipline. It asks it to be measured against a different question: not how much was written, but how reliably the writing can be repeated by the system now standing between a brand and its buyer.

Where does this leave media relations?

Closer to the center of the work than it has been in years, on the current evidence.

For most of the past decade, the pressure ran toward owned channels — the blog, the funnel, the content a brand fully controlled. The citation data complicates that drift. The surfaces a company controls least, the independent ones, are the surfaces an engine appears to trust most.

That does not retire owned content; a brand still needs a clear page stating its own facts for the coverage to confirm. But it does reposition earned media from a nice-to-have into the part of the mix that a machine treats as evidence. Run owned and earned together long enough and they begin to reinforce each other — a quiet, compounding effect that a coordinated approach to a brand’s narrative is built to produce. The label matters less than the finding underneath it: in the AI era, the press is reading on your buyer’s behalf, and PR is how a brand earns a place in what it reads.

Frequently asked questions