A series of 2025 and 2026 citation studies has produced one of the cleanest findings in the history of digital communications research. Earned media, not owned content, is the channel that compounds inside AI answers. The research is consistent across every major engine.
A series of citation studies published between October 2025 and April 2026 has produced one of the cleanest findings in the history of digital communications research. Across ChatGPT, Perplexity, Gemini, and Claude, AI answer engines do not trust what brands say about themselves.
Muck Rack's December 2025 analysis of generative AI citations found that 94% of AI citations came from non-paid, non-brand-owned sources. A controlled experiment by University of Toronto researchers, referenced extensively across the GEO research community, confirmed the bias is structural rather than incidental: AI search engines show systematic, overwhelming preference for earned media over brand-owned content and over social content. The researchers' direct conclusion was that brands "must dominate earned media to build AI-perceived authority." A study referenced across multiple GEO research firms found that 82% to 89% of AI-generated answers cite earned media rather than brand websites or blogs.
The supporting datasets sharpen the picture.
Stacker's December 2025 analysis found that distributing content across a wide range of publications — rather than publishing only on owned channels — increases AI citations by up to 325%. Ahrefs' analysis of 75,000 brands found that branded web mentions correlate three times more strongly with AI visibility than backlinks do. The volume of times a brand is mentioned across the open web in third-party context predicts AI visibility better than any traditional SEO authority signal.
The platform-by-platform variation matters because the prescription differs by engine.
Perplexity is the platform where the earned media gap is most visible. Yext's analysis of 6.8 million citations across Gemini, ChatGPT, and Perplexity, published October 2025, found Perplexity sources narrowly into industry-specific publications and review platforms — Zocdoc in healthcare, TripAdvisor in hospitality, Reddit broadly. A separate analysis from Discovered Labs found Reddit accounts for 46.7% of Perplexity's top 10 citations. Brand-owned domains rarely earn direct Perplexity citations regardless of how strong the on-page SEO investment is. AuthorityTech's research on Perplexity's three-layer reranking system found the platform's ranking architecture structurally favors earned media from Tier-1 publications because of how its authority signals interact with externally verified credibility cues. A Forbes article about a company has passed an editor's judgment. The company's own blog post has not. Perplexity's reranker reads the difference.
Gemini is the inverse. Yext's data shows 52.15% of Gemini citations come from brand-owned websites with structured data and consistent subdomains. Gemini trusts what a brand says — provided the brand says it inside the schema and entity-definition framework Google has spent two decades training the index on. For Gemini, the prescription is technical SEO discipline, schema completeness, and entity consistency across the Google ecosystem.
ChatGPT lands in the middle. Heavy reliance on third-party directories and listing platforms (48.73% of citations per Yext) and a strong concentration on Wikipedia (47.9% of ChatGPT's top-10 citations per Discovered Labs) mean the prescription is broad listing-platform coverage and encyclopedic entity definition. Earning a Wikipedia mention dramatically increases ChatGPT citation likelihood. So does coverage in Yelp, BBB.org, TripAdvisor, and category-specific listing platforms.
Claude sits at the most cautious end of the spectrum. It cites less frequently than ChatGPT or Perplexity, prioritizes technical precision, and shows the strongest preference for structured, sourced, authoritative content. AI traffic from Claude converts at 5% according to Position Digital's April 2026 compilation. Lower volume than ChatGPT, but a high-quality consideration audience for B2B and professional categories.
The takeaway for any communications team operating a 2026 budget is that the channel that has compounded value across all four major answer engines is earned media. Press release distribution is part of it — PR Newswire surfaces with surprising regularity in time-sensitive coverage on ChatGPT and Google AI Overviews — but it is not the whole of it. The engines reward bylined editorial coverage, analyst-grade independent research, structured review platform presence, and high-credibility category publications. The brands that build that coverage at scale, and at consistency, are showing up in answers. The brands relying on owned content alone are not.
There is a 70-year-old argument inside the communications industry about whether earned media is structurally more valuable than paid or owned. The AI citation data has now resolved that argument empirically, in a channel that did not exist when the argument started. The implication for budget allocation is that the channels which were already underweighted relative to their long-term ROI — earned media, analyst relations, third-party research, executive thought leadership — are the channels that compound inside the new search layer. The channels that were overweighted relative to long-term ROI — programmatic display, owned content production at scale, owned social — do not.
The agencies that can build earned media authority and measure it inside answer engines are now selling a service that the data shows works. The agencies that cannot are selling against a measurement system that no longer reflects how brand authority gets built.
Sources
Position Digital — 150+ AI SEO Statistics for 2026 (compiles Muck Rack, Ahrefs, Stacker)
AuthorityTech — How to Dominate GEO with Earned Media Citations in 2026
AuthorityTech — How Perplexity Selects Sources
Yext — AI Visibility in 2025: How Gemini, ChatGPT, and Perplexity Cite Brands
Discovered Labs — AI Citation Patterns
Profound — AI Platform Citation Patterns
