Earned Media

Earned media is coverage of a brand that the brand neither pays for nor controls: press articles, independent reviews, “best of” lists, forum threads, third-party videos and encyclopedia entries. It is the third leg of the long-standing paid, owned and earned model. In AI search it has become the leg that carries most of the weight, because the sources answer engines cite are predominantly third-party. The size of that majority depends heavily on how it is measured, and the evidence that earned coverage causes AI visibility is weaker than the evidence that the two go together.

Third-Party Versus Owned Citation Shares

Several independent datasets agree on the direction. Researchers at the University of Toronto (arXiv, September 2025) found AI search engines citing earned media 72 to 92% of the time against 18 to 27% for brand-owned content. Żatuchin (arXiv, June 2026; about 168,000 citations, 128 brands, 12 markets) found 85.7% of citations pointing to sites the brand does not own and 14.3% to owned sites. Kumar (arXiv, June 2026; a vendor-authored study of 149,912 citations across five engines) put the brand-owned domain at just 2.9% of citations.

The spread between 2.9% and roughly a quarter is not noise. It reflects different prompts, engines and definitions of “owned.” Owned pages do much better when the question is about a specific product. Ten Speed (September 2026; vendor study, 7,387 citations from 220 B2B evaluation-stage prompts) found product pages were the single largest category at 24.1%. In local search, Yext (October 2025; 6.8 million citations) found first-party websites supplying 44% of citations and listings another 42%. The fair summary is that earned media dominates open-ended category questions, while owned pages hold their ground once a buyer names the product or the place.

Listicles and Review Sites

Two earned formats stand out. Kumar found “best-of” listicles accounting for 21.0% of all citations, the largest single page type in that dataset. In Ten Speed's B2B sample, comparison content (13.3%) and listicles (13.2%) together outweighed product pages, with G2 and Capterra at 7.2% and community sites at 4.2%. Whitespark's 2026 Local Search Ranking Factors, an opinion survey of 47 practitioners and not a measurement, ranked presence on expert-curated “best of” lists as the top factor for AI visibility.

Buyers appear to share the engines' preference. In G2's April 2026 survey of 1,076 B2B software decision-makers, 45% named review-site citations as the most confidence-inspiring signal in an AI response. G2 is a review site, so the finding is self-interested, but it is consistent with the citation data. Community platforms sit in the same bucket: Reddit, YouTube and Wikipedia lead Ahrefs' September 2026 lists of the most-cited domains in ChatGPT, Perplexity and Google AI Overviews.

Traditional SEO treated earned media mainly as a source of backlinks. For AI visibility, the mention itself appears to matter more. Ahrefs' study of 75,000 brands (December 2025) found brand visibility in AI answers correlated most strongly with YouTube mentions (about 0.71 to 0.74 across ChatGPT, AI Mode and AI Overviews) and branded web mentions (0.66 to 0.71). Backlink counts (about 0.25 to 0.28) and Domain Rating (0.27 to 0.33) trailed well behind.

One plausible mechanism is that language models learn associations between a brand and a category from how often the two appear together in text, whether or not a hyperlink is present. The NanoKnow study (SIGIR 2026) found models “more accurate when answers were seen during pre-training,” an advantage that retrieval reduces but does not remove. This is the link between earned media and training data frequency.

Caveats

The Ahrefs figures are correlations. Large, well-known brands attract more mentions, more videos and more AI visibility at once, and no public study has isolated the effect of adding earned coverage to an otherwise unchanged brand. Kumar's day-one visibility figures (73% for household brands, 11% for niche brands) suggest much of the advantage is simply existing fame.

Earned sources are also unstable and open to abuse. Provisional Promptwatch data reported by Semrush showed Reddit's share of all ChatGPT citations falling from 3.8% to 0.5% in mid-August 2026, so a strategy resting on one platform inherits that platform's volatility. Because listicles carry so much weight, they are a target for manipulation: the FORGE study (arXiv, June 2026) found a single polluted page could sway up to 27% of recommendations. Engines have an incentive to discount self-serving lists, and the GEO Defender work (September 2026) shows defences that cut attack success from about 50% to 6%. Coverage that is earned in the ordinary sense is the kind most likely to survive that filtering. Whether a given programme is working is a question for AI visibility measurement.

Further Reading