Two pieces published within a day of each other describe the same crisis from opposite ends of the market, and neither names it. One reports that Google search traffic to publishers has fallen by a third in a year. The other tells communicators that earned media in AI Search now includes editorial content they paid to place. Read together, they tell the real story: the open web's payment rail broke, and the industry is improvising a replacement out of PR.

  • Chartbeat data in the Reuters Institute's 2026 trends report shows Google organic search traffic to more than 2,500 news sites fell 33% globally between November 2024 and November 2025, and 38% in the United States.
  • With the traffic rail gone, the PR industry is quietly widening the definition of earned media to cover paid editorial placement, which is the agentic web's version of paid inclusion.
  • Citation data does not reward that tactic: Muck Rack's Generative Pulse analysis of more than 25 million cited links puts earned editorial near 84% of AI citations and press releases below 2%.
  • The missing layer is not more content. It is a labeled, measured ad unit inside the answer that pays the publisher whose content grounded it.

What earned media in AI Search actually means now

Earned media in AI Search is a third-party mention that an AI engine chooses to cite when building an answer, without the brand having bought the placement. The distinction matters because the alternative already has a name from the last cycle. Paid inclusion was the practice of paying to enter an index or a result set with no disclosure label, and US regulators pushed search engines toward clear labeling of paid placement in 2002. The agentic web has arrived at the same junction roughly twenty-four years later, with no labels, no auction, and no split.

From the link economy to the citation economy: who gets paid

In the link economy the loop closed. Publishers made content, Google sent a click, an advertiser paid for the click, and a share of that money reached the publisher. In the citation economy the loop is open. The engine ingests the content, synthesizes the answer, cites the source, sells the placement beside it, and keeps the proceeds. The mechanic changed. The obligation did not.

What the two sources reported, and what we verified

The Google Zero piece is directionally right and numerically loose. Its own summary claims AI Overviews appear on 48% of searches. The SparkToro study it cites, published in June 2026 on Similarweb clickstream data, reports AI Overviews on more than 20% of Google searches, with click-through falling roughly 60% when they appear. We use SparkToro's numbers, including the headline: 68.01% of US Google searches ended without a click in the first four months of 2026, up from 60.45% in 2024.

The earned media piece has the same problem in the other direction. It attributes to Bain the claim that roughly 70% of consumers use AI tools in product research. Bain's own April 2026 buyer research puts it differently: 44% of US online buyers mostly start in an LLM or split between AI and classic search. L.E.K. Consulting, also April 2026, finds about 30% have used AI to inform a purchase decision. The direction is settled. The magnitude is not, and inflated figures make budget conversations easier to win and much harder to keep.

Why a bigger content footprint is not an AI visibility strategy

The strongest version of the PR industry's case

The reframe is not cynical, and it is partly right. AI engines do weigh authority, consistency, recency and frequency, and the disciplines that manufacture those signals live in communications teams, not in performance marketing. Meltwater's GenAI Lens research across more than eight million citations in eight major models found earned and news media accounted for 37.6% of citations in May 2026. If credibility is the currency, communicators hold the mint.

Where it breaks: bought volume does not convert into citation

The evidence on distributed content is unambiguous. A BuzzStream analysis of four million AI citations found syndicated press releases on platforms such as Yahoo Finance and MSN accounted for 0.04% of citations, and wire domains 0.21%. Meltwater tracked press release citations falling from 0.4% in April 2026 to 0.2% in May. Identical copy pushed across thousands of sites reads to a crawler as duplication, not consensus. And it pays the distribution network rather than the publishers whose reporting the engine actually cites.

Calling a paid placement earned media does not persuade an engine to cite it. It only makes the invoice harder to audit.

— Smalk AI

What this means for brands and for publishers

For CMOs, media buyers and agencies: buy the label, not the ambiguity

  • Split the budget lines. PR earns citations, media buying purchases labeled placement inside answers. Blending them guarantees you never learn which one worked.
  • Demand answer-level measurement. OpenAI's public ad documentation describes cost-per-click buying and aggregated reporting, not attribution to an individual prompt or conversation stage.
  • Audit your category this quarter. Ask an assistant the five questions your buyers ask, log who gets cited, and treat that list as your real share of voice.

For publishers: sell the citation, not the syndication slot

Licensing will not reach most publishers, and revenue share is not on the table where the money is. OpenAI's VP of media partnerships told the WAN-IFRA congress in Marseille in June 2026 that the company has no plans to share ChatGPT advertising revenue with publishers, including those holding licensing deals. The Reuters Institute meanwhile found platform funding ranked as the top new growth opportunity for 2026, while only about a fifth of media leaders expect substantial AI platform revenue this year. That gap is the business. Price inventory on citation, and sell it through a channel the long tail can actually access.

Three signals that decide who gets paid by 2028

Three things are already in motion. Demand is proven: OpenAI launched ads in ChatGPT in February 2026 and reported 100 million dollars in annualized revenue within six weeks, with more than 600 advertisers. Supply is unresolved: Perplexity shared ad revenue with publishers and then pulled advertising over trust concerns, ProRata has committed to a 50% split, and eMarketer still expects US chatbot ad revenue below one billion dollars in 2026. Disclosure is untouched: sponsored content inside a synthesized answer has no agreed label, which is exactly the condition regulators acted on last time. Whoever standardizes the label and the split will set the price for everyone else.

Conclusion

Hold on to this: the collapse of search traffic did not create a shortage of content, it created a missing payment rail, and earned media in AI Search cannot be stretched to cover it. Smalk AI is building that rail as generative engine advertising: native ads placed inside AI answers, labeled as advertising, measured where the answer happens, with a revenue stream for the publishers whose content grounds the response. Watch for the first standardized disclosure format for sponsored citations, because whoever sets the label will also set the split.