# Earned Media vs Owned Media > Earned media vs owned media in AI citations: why studies put the brand-owned share anywhere from 2.9% to 44%, and how to split budget between the two. Source: https://metavert.io/compare/earned-media-vs-owned-media Published: 2026-10-07 Updated: 2026-10-07 Comparison **Earned media** is coverage a brand does not control, such as reviews, press, community threads, third-party lists and videos; **owned media** is what the brand publishes on properties it controls, such as its website, documentation and product pages. In [AI search](https://metavert.io/ai-search), the split between [earned media](https://metavert.io/earned-media) and owned media decides whose words an assistant repeats when someone asks about a category. The studies do not agree on that split, and the disagreement is informative. Kumar (June 2026) puts brand-owned domains at 2.9% of AI citations. Ten Speed (September 2026) finds product pages alone are 24.1% of citations for evaluation-stage B2B prompts. Yext (October 2025) finds first-party websites are 44% of citations in local search. All three can be right, because each divides by a different set of questions. The consistent pattern underneath is that earned sources dominate when a buyer is discovering a category, and owned sources gain ground as the question becomes specific to a product or a place. ## Feature Comparison | Dimension | Earned Media | Owned Media | | --- | --- | --- | | What it is | Reviews, press, community posts, third-party lists, others' videos | Brand website, product pages, documentation, brand channels | | Control | Low: someone else writes and hosts it | High: the brand writes, updates and structures it | | Share of citations, cross-engine | The large majority of 149,912 citations (Kumar, Jun 2026) | 2.9% on brand-owned domains | | Share of citations, 12 markets | 85.7% on sites the brand does not own (Żatuchin, Jun 2026) | 14.3% | | Share of citations, earlier estimate | 72% to 92% (GEO Toronto, Sep 2025) | 18% to 27% | | Share, B2B evaluation-stage prompts | Comparison content 13.3%, listicles 13.2%, G2/Capterra 7.2%, community 4.2% (Ten Speed, Sep 2026; mix of owners) | Product pages 24.1% | | Share, local queries | Reviews and social 8%; listings 42% (Yext, Oct 2025) | First-party websites 44% | | Strongest role | Getting a brand into the answer at the discovery stage | Supplying accurate specifics once the brand is being evaluated | | Correlation with AI visibility | YouTube mentions 0.71 to 0.74; web mentions 0.66 to 0.71 (Ahrefs, Dec 2025) | Not measured directly in that study | | Buyer trust | 45% of B2B buyers call review-site citations the most confidence-inspiring signal (G2, Apr 2026) | Authoritative for facts about itself | | Speed and cost | Slow, relationship-driven, uncertain | Fast, predictable, fully in the team's hands | | Main risk | Coverage that is absent, outdated or wrong, with no edit rights | Being accurate and uncited | ## Detailed Analysis ### Why the Owned Share Ranges From 2.9% to 44% The three headline figures measure different things. Kumar's 2.9% comes from 149,912 citations across five engines (the author works for a vendor), in a sample where “best-of” listicles alone account for 21.0% of all citations. Prompts that ask which products exist or which is best get answered from lists and roundups, and a brand's own site is one voice among many. Ten Speed's 24.1% comes from 7,387 citations on 220 evaluation-stage B2B prompts (also vendor research). At that stage the buyer is already weighing named products, so the engine needs pricing, features and integrations, and the product page is the obvious place to find them. The figure is for product pages as a page type across the vendors in the answer, not a single brand's share of its own mentions. Yext's 44% comes from 6.8 million citations for local queries. A question about a particular business's hours, services or location has one canonical source, which is that business. The next 42% went to listings, a category that sits between the two definitions: hosted by a third party and largely filled in by the business. So the denominator changes three ways at once: the stage of the question, the vertical, and what counts as owned. Żatuchin's 14.3% across 128 brands and 12 markets, and the 18% to 27% reported by the GEO Toronto paper in 2025, fall between the extremes for the same reason. None of these numbers is a universal owned-media share, and quoting one without its prompt set is misleading. ### What Earned Media Does That Owned Cannot Earned coverage decides whether a brand is considered at all. Ahrefs' study of 75,000 brands (December 2025) found the strongest correlates of AI visibility were brand mentions on YouTube and across the web, well ahead of backlinks or Domain Rating. Those are correlations, not proof of cause. Kumar's data shows the head start they imply: day-one visibility was 73% for household brands, 44% for mid-market and 11% for niche brands. Part of that advantage is baked in before retrieval, since models are more accurate about entities they saw often in training (NanoKnow, SIGIR 2026), a point developed under [training data frequency](https://metavert.io/training-data-frequency). Concentration raises the stakes. Żatuchin found about 80% of citations come from roughly 18% of domains, with [Wikipedia](https://metavert.io/wikipedia) the top source in 11 of 12 languages. Even on a brand's own social platform the earned pattern holds: Goodie (September 2026) found roughly nine in ten cited LinkedIn URLs sit outside the company page. In local search, a Whitespark survey of 47 practitioners (November 2025; expert opinion, not measurement) ranked presence on expert-curated “best of” lists as the top AI-visibility factor. ### What Owned Media Does That Earned Cannot Owned pages are where a brand can be exact and current. Seer Interactive (July 2026) found 75% of pages cited by language models had been updated within the last year, and only the owner can guarantee that for product facts. When an engine cites a stale third-party review for pricing, the remedy is usually a clearer, fresher first-party page for it to prefer. Owned media also matters most where money changes hands. The Ten Speed and Yext figures both describe moments close to a decision. A brand that has earned its way into the shortlist and then presents thin or contradictory product pages hands the detail of the answer to whoever else wrote about it. The Whitespark panel's second-ranked factor was a dedicated page for each service. ### Risk and Measurement Dependence on third parties has a cost. Research on manipulation (FORGE, 2026) found a single polluted page could sway up to 27% of recommendations in its tests, and a brand has no edit rights over what others publish. Sources also rotate: day-to-day similarity of cited sources averages 0.34 to 0.42 (Schulte, Bleeker and Kaufmann, April 2026). The useful measurement is therefore the owned-versus-earned mix for a brand's own prompts, split by funnel stage and tracked over time, which is the kind of breakdown [AI citation](https://metavert.io/ai-citations) tools such as [LLM Optimizer](https://metavert.io/llmopt) are built to show. An industry average will not reveal whether a particular category behaves like Kumar's sample or like Ten Speed's. ## Best For #### If you can only fund one this quarter and assistants do not mention you Earned media Absence at the discovery stage is a third-party coverage problem. More pages on your own site will not put you on lists and in reviews that engines draw from. #### If you can only fund one and assistants mention you but get the facts wrong Owned media Fix and refresh product, pricing and documentation pages so there is a current first-party source to cite. This is fast and entirely within your control. #### If you sell B2B software to buyers comparing named vendors Both, owned first Product pages were the largest single citation type in evaluation prompts (Ten Speed, September 2026), with comparison content, lists and review sites together larger still. #### If you are a local business Owned media First-party sites and listings made up 86% of local AI citations in Yext's data. Get the website and listings complete and consistent before pursuing press. #### If you are a niche or new brand Earned media Day-one visibility for niche brands was 11% in Kumar's data against 73% for household names. The gap is made of mentions you do not yet have. #### If leadership asks for the “right” owned share Depends on the prompt set Published figures range from 2.9% to 44% because they measure different questions. Measure your own prompts by stage instead of adopting a benchmark. #### If you are choosing between a blog programme and a review-site programme Earned media G2's 2026 survey found review-site citations are the signal B2B buyers trust most in an AI answer. Reviews serve both the engine and the reader. #### If your earned coverage is strong but ageing Owned media You cannot update other people's articles. Keeping your own pages current gives engines a fresher alternative to old third-party claims. ## The Bottom Line Earned and owned media do different jobs in AI answers, and the conflicting statistics reflect that. Across broad, discovery-style prompts, citations go overwhelmingly to sites a brand does not own. As questions narrow to a named product or a specific location, first-party pages take a much larger share. A figure for the owned share means little without the prompts it was measured on. For most brands the order is earned first, owned close behind. Third-party mentions are the best-supported correlate of being included at all, and they are the slower asset to build. Owned pages then determine whether the details in the answer are accurate and current, which only the brand can ensure. The practical step is to measure the mix for your own category and stage of funnel before setting a budget split. The published ranges are wide enough that any single benchmark could be wrong for a given business by an order of magnitude. ## Related Topics - [Earned Media](https://metavert.io/earned-media) - [AI Citations](https://metavert.io/ai-citations) - [AI Share of Voice](https://metavert.io/ai-share-of-voice) - [Training Data Frequency](https://metavert.io/training-data-frequency) - [Generative Engine Optimization](https://metavert.io/generative-engine-optimization) - [Local AI Search](https://metavert.io/local-ai-search) - [Wikipedia](https://metavert.io/wikipedia) - [Reddit](https://metavert.io/reddit) - [LLM Optimizer](https://metavert.io/llmopt) ## Further Reading - [Kumar, 149,912 AI Citations Across Five Engines (June 2026) – arXiv](https://arxiv.org/html/2606.20065) - [What AI Cites at the B2B Evaluation Stage (September 2026) – Ten Speed](https://www.tenspeed.io/blog/what-ai-cites-b2b-evaluation-stage) - [AI Citations Study, 6.8M Citations (October 2025) – Yext](https://www.yext.com/about/news-media/ai-citations-release) - [Żatuchin, Owned vs Non-Owned Citations Across 12 Markets (June 2026) – arXiv](https://arxiv.org/abs/2606.25787) - [Generative Engine Optimization: How to Dominate AI Search (September 2025) – arXiv](https://arxiv.org/abs/2509.08919) - [AI Brand Visibility Correlations, 75K Brands (December 2025) – Ahrefs](https://ahrefs.com/blog/ai-brand-visibility-correlations) - [NanoKnow – SIGIR 2026 / arXiv](https://arxiv.org/abs/2602.20122) - [Social Media AI Citations Study (September 2026) – Goodie](https://higoodie.com/blog/social-media-ai-citations-study-2026/) - [Local Search Ranking Factors 2026 – Whitespark](https://whitespark.ca/local-search-ranking-factors/) - [Half of B2B Software Buyers Now Start Research With AI Chatbots (April 2026) – G2 via PR Newswire](https://www.prnewswire.com/news-releases/new-g2-research-half-of-b2b-software-buyers-now-start-their-research-with-ai-chatbots-302742807.html) - [Content Recency's Impact on AI Visibility (July 2026) – Seer Interactive](https://www.seerinteractive.com/insights/study-content-recencys-impact-on-ai-visibility-in-2026) - [FORGE: Manipulating AI Recommendations (2026) – arXiv](https://arxiv.org/abs/2606.13610) - [Schulte, Bleeker and Kaufmann, Don't Measure Once (April 2026) – arXiv](https://arxiv.org/abs/2604.07585)