Earned Media Is Now AI Infrastructure

The ten blue links are going away.

Start with the numbers, because the numbers are startling. Google AI Overviews now appear on 48% of U.S. searches and reach two billion monthly users. Google AI Mode has 75 million daily actives. ChatGPT went from 400 million to 900 million weekly users in twelve months. Gemini hit 750 million MAU. And 69% of all searches now end without anyone clicking anything, up from 56% a year ago.

That last number is the one that matters most. When a search ends without a click, the only brand impression that gets delivered is the one inside the AI-generated summary. There is no landing page. There is no retargeting pixel. There is no funnel. There is only: were you mentioned, or weren't you?

So where do AI engines get the brands they mention? Muck Rack analyzed over a million AI-generated responses in December 2025 and found that 82% of cited links came from earned media. 95% came from non-paid coverage of any kind. A separate academic study by Fullintel and the University of Connecticut, presented in February 2026, found 89% earned media citation rates, with 47% coming from journalistic sources specifically.

Two independent studies. Different methodologies. Same conclusion. When AI engines recommend brands, they are overwhelmingly drawing from editorial coverage, not from brand-owned content, not from paid placements, and not from SEO-optimized landing pages.

SEO does not predict AI citation.

That last bar in the chart deserves its own section, because it overturns a core assumption. Only 12% of URLs cited by ChatGPT, Gemini, and Copilot rank in Google's top 10 for the same query (Ahrefs). Moz ran 40,000 queries through Google's AI Mode and found that 88% of the citations were not in the traditional organic top 10.

Read that again. A brand can own its target keywords in Google and be completely absent from AI answers in its category. The two systems pull from different signals. Traditional search rewards on-page optimization, backlink profiles, and technical SEO. AI engines reward something else: whether independent, credible publications have written about you, recently, and with specificity.

The top 10 domains capture 46% of all ChatGPT citations on a given topic. Wikipedia alone accounts for 7.8%. Reddit citations grew 450% in four months. LinkedIn dominates professional queries. The concentration is extreme, and the common denominator across all of these is third-party credibility: someone other than the brand, saying something about the brand, in a venue the AI engine trusts.

The SEO playbook does not transfer to AI visibility. Different system, different inputs, different winners. The input that matters most is whether a credible publication has written about you.

Almost every brand is recognized when you ask by name. Almost none are recommended unprompted.

There is one data point from the research that I keep coming back to, because it captures the entire problem in a single ratio.

When AI systems were asked about a specific product by name, recognition rates hit 99.4%. The AI knows the brand exists. But when asked category-level discovery questions, the kind that actually drive purchase decisions ('What are the best tools for X?'), discovery rates on ChatGPT dropped to 3.32%.

A 30-to-1 gap between being known and being chosen.

The study found zero correlation between GEO content optimization scores and discovery rates. The variable that predicted whether a brand got recommended was referring domains and third-party editorial presence. Not how well the brand's own pages were optimized. Not keyword density. Not schema markup. Whether other people, writing in other publications, had written about the brand with enough consistency that the AI engine treated it as a credible answer.

This is the data point that changes the budget conversation. For executives who have always been skeptical of PR, who have always wanted harder ROI attribution, who have always treated earned media as a nice-to-have: the question is no longer 'does brand awareness matter.' The question is 'does your brand appear when someone asks an AI engine for a recommendation in your category.' And the answer to that question is determined, overwhelmingly, by earned media.

The money is moving. Fast.

Gartner's 2026 communications predictions included a headline number: by 2027, mass adoption of public LLMs as a replacement for traditional search will drive a 2x increase in PR and earned media budgets. Double. Their reasoning is structural, not aspirational. More than 95% of links cited by AI engines are non-paid. Traditional SEO is best served by marketing functions. Answer engine optimization requires skills that live in communications teams.

The CEO survey data from Delight Labs, published March 2026, is more granular. 92% of CEOs surveyed have increased PR investment specifically because of AI-powered search engines. Not generally. Not because of a vague sense that PR matters. Because of AI search. 49% described the increase as significant or extensive. When asked which tactic they believe most drives AI citation, earned media came first (29%), ahead of technical SEO (22%).

Forrester's 2026 B2B data adds a telling contrast. 83% of marketing decision-makers expect total marketing spend to rise. But content and creative services? Declining. From 53% planning increases in 2024 to 44% in 2025. The budget is not growing across the board. It is migrating from the content that AI can produce for itself toward the placements that AI cannot generate: third-party editorial coverage.

The reason is embarrassingly simple.

AI engines were trained on the internet. The internet's most trusted information lives in publications with editorial standards. When someone asks an AI 'who leads this category,' the model does not crawl your homepage and evaluate your claims. It synthesizes from the sources it learned to trust during training, and retrieves from during inference. 65.3% of pages cited by ChatGPT come from domains with a domain rating of 80 or higher (Ahrefs). The AI is, in effect, checking references.

The Princeton and Georgia Tech research on Generative Engine Optimization, published at SIGKDD 2024, quantified this. Adding statistics from credible sources to content improved AI citation rates by 30 to 40%. Adding source citations alone delivered a 41% visibility gain. The AI does what a good analyst does: it weights the source, then it extracts the claim.

Earned media produces 325% more AI citations than owned content making the same claim (AuthorityTech). That is not a marginal advantage. It is the difference between a system that trusts third parties and one that does not trust self-reported information. Which, when you think about it, is exactly how human credibility works too.

One more wrinkle. Recency matters significantly. Over half of all AI citations come from content published in the last twelve months. Citation rates peak within seven days of publication (Muck Rack). 70% of AI Overview citations rotate out within two to three months (5WPR). This is not a project. It is a program. The closest analogy is not building a website. It is running a press office.

The reader changed. The mechanism did not. Third-party credibility in a trusted publication has always carried more weight than what a brand says about itself. That was true when the reader was a journalist. It was true when the reader was a buyer. It is true now that the reader is a machine.

As of two days ago, this has its own measurement standard.

On May 20, AMEC, the organization responsible for the Barcelona Principles (the communications industry's accepted measurement framework), published GEO Principles and a Practitioner's Guide to GEO Measurement. Three areas: upstream reputation signals like earned coverage and expert commentary; search and content readiness, meaning whether a brand's presence is structured for AI interpretation; and downstream AI outputs, meaning how the brand actually appears in AI answers, including citations, framing, and conspicuous omissions.

Their guidance is measured. AI outputs are directional evidence, not ground truth. No single score or platform should be relied on in isolation. The most useful measurement triangulates across all three layers. This is sound advice, and it is also a meaningful institutional signal: the body that wrote the global standard for communications measurement now recognizes AI visibility as a formal measurement domain.

Gartner predicts 45% of Chief Communications Officers will adopt narrative intelligence technologies by 2029 to track how brands are represented across media, search, social, and AI-generated answers. CCOs forecast their tech investments will grow faster than any other spend category. The infrastructure is being built.

Three things follow from the data.

First, for agencies: your core competency just became the most strategically valuable skill in the marketing stack, and that is not hyperbole. Pitching credible stories, building editorial relationships, earning the kind of coverage that a skeptical reader would take seriously. These skills are what determines whether a client appears in AI-generated recommendations. The agencies that build GEO measurement, strategy, and ongoing programs around earned media will define the category. The ones that treat it as a media monitoring add-on will cede it to someone else.

Second, for in-house comms leads: 92% of CEOs in the Delight Labs survey already connected earned media to AI visibility. If your leadership team has not, the data now makes the argument. The conversation is no longer 'PR is hard to measure.' It is: 'PR determines whether we appear when someone asks an AI for a recommendation, and we can now measure citation frequency, citation share, and recommendation probability across ChatGPT, Perplexity, Gemini, and Google AI Overviews.'

Third, for everyone: PR, search, and social are no longer separate functions and they cannot be run as separate programs. Credibility gets established through earned media. It gets amplified across social and owned channels. And it gets retrieved by AI engines during inference. One system. The brands that build all three layers will be the ones that show up in AI answers. The brands that optimize only one will wonder why they are invisible.