Earned media is a meaningful source class to measure when buyers use ChatGPT, Perplexity, Gemini, and other answer systems for category research. Recent studies have found substantial representation of editorial, journalistic, earned, and non-paid sources within sampled AI citations. Those observations justify measuring whether independent coverage appears in the answers that matter to a brand. They do not reveal one universal source-selection mechanism or prove that earning a placement will cause a citation, recommendation, or sale.
Forrester's State of Business Buying 2026 reported that 94% of surveyed B2B buyers use AI during their purchase process. That buyer behavior makes AI-mediated discovery a practical communications concern: teams need to know what claims, entities, publications, and URLs appear when buyers ask category and vendor-comparison questions.
Key Takeaways
The Muck Rack Generative Pulse analysis examined more than one million cited links in sampled AI responses as of December 2025. Within that measured sample, 82 percent of cited links came from sources the brand neither owned nor paid for, about 94 percent were classified as non-paid — a wider category that still includes brand-owned content — and press releases represented 6 percent in the report's press-release category. The May 2026 edition put the non-owned, non-paid share at 84 percent. It also reported a recency distribution within the citations it observed.
The measured unit is the composition and age of links already present in the sampled responses. It does not establish that earned coverage or recency caused selection, that engines use a fixed source hierarchy, or that a specific publication or press release will be retrieved, cited, or recommended.
The Fullintel-UConn study presented at IPRRC measured source types in AI responses to a health-focused set of weight-loss-drug queries — 400 prompts across 10 personas, run on a single platform. Third-party news and informational sources represented 47 percent of cited links in that sample; corporate, university and health-network sites represented 48 percent.
These are source-composition observations within the study's sampled responses. They do not establish a universal engine-selection mechanism, prove that journalism or earned media is the primary input, or show that coverage causes a brand citation, recommendation, referral, conversion, pipeline result, or revenue.
The Ahrefs analysis of ChatGPT's most-cited pages grouped pages that had already appeared as citations by Domain Rating. It reported that 65.3 percent of the cited pages in its bounded inventory came from domains rated 80 or above.
The measured unit is the domain-rating distribution of pages that had already been cited, not a domain-rating mechanism. The inventory does not establish that a high Domain Rating caused citation, that DR80+ should be a universal publication-selection threshold, or that a placement on a high-rated domain will produce a brand mention or recommendation.
AuthorityTech's earned-versus-owned analysis observed a 4.25× citation-rate difference between earned and owned records across the topics it compared. The measured unit is a bounded observed rate in that dataset. It is not evidence of causality or primacy, does not recommend one distribution strategy for every brand, and is not a forecast, guarantee, or business outcome.
The useful bridge to Machine Relations is measurement: compare source role, placement, retrieval, citation, recommendation, referral, and commercial outcome separately. A placement is a communications output. Retrieval, citation, recommendation, and revenue are distinct downstream observations that require their own evidence.
Independent coverage can add a third-party, attributable account of a company, category, product, dataset, or point of view. That makes it useful to human buyers and creates another public artifact that an AI system may be able to access. The defensible claim is that the artifact exists and can be tested—not that publication guarantees machine use.
The practical value of earned media comes from roles that can be inspected directly:
These roles make independent coverage a useful input to an AI-visibility program. They do not make earned media sufficient. Technical access, owned information, entity consistency, query relevance, engine-specific retrieval, and answer construction can all affect what appears.
Platform-level citation behavior changes by product, query, geography, freshness, index, and interface. A current answer from one engine is not a timeless rule for every engine. Instead of publishing a single trust hierarchy, build a dated query panel and inspect each answer as an observation.
For every engine and query, record:
The Yext AI Citation Refresh analyzed 17.2 million citations across multiple AI engines and reported platform-specific source patterns. The OtterlyAI AI Citations Report 2026 likewise describes a large observational corpus. These inventories can help teams choose dimensions to measure, but they do not supply a permanent engine-wide rule for a given brand.
An evidence-led earned-media strategy begins with buyer queries, publication relevance, and a measurement plan. It does not begin with a universal Domain Rating threshold or a promise that editorial placement will translate automatically into an AI citation.
Step 1: Establish a dated answer baseline. Run the category, comparison, problem, and vendor questions buyers use across the engines relevant to the business. Save answer text, brand mentions, cited hosts, exact cited URLs, and recommendation language. AuthorityTech's AI visibility audit provides a starting measurement surface.
Step 2: Map source roles, not just publication scores. Identify which editorial, analyst, academic, community, government, directory, press-release, and owned sources appear in the answer set. Use topical relevance, editorial fit, audience, access, and observed query coverage to prioritize publications. Do not treat Domain Rating as a disclosed engine-selection formula.
Step 3: Develop evidence worth publishing. Original data, named methodology, attributable expertise, clear definitions, and specific examples make a story more useful to an editor and easier to evaluate later. The Aggarwal et al. GEO benchmark supports testing sourced and attributable interventions under stated conditions; it does not promise that adding a statistic will secure publication or citation.
Step 4: Earn relevant editorial coverage. Coverage should be evaluated as an editorial artifact with its own audience and purpose. Muck Rack's observed sample supports distinguishing editorial, non-paid, paid, press-release, and owned source roles. It does not show that paid material is never cited or that editorial status alone produces selection.
Step 5: Re-run the query panel. After publication and indexing, test the same queries on a defined cadence. Separate publication, accessibility, retrieval, host citation, exact-URL citation, attributed claim, recommendation, referral, and commercial outcomes. A change after publication is an observation to investigate, not automatic proof of cause.
These disciplines address different parts of machine-mediated discovery. Machine Relations provides the broader measurement architecture:
| Discipline | Primary object | Observable result | Boundary |
|---|---|---|---|
| SEO | Search crawling, indexing, and ranking | Indexed URL, ranking, impression, click | A ranking does not guarantee inclusion in an AI answer |
| GEO | Generative answer surfaces | Visibility, citation, attributed claim | A content intervention does not guarantee live-engine selection |
| AEO | Direct-answer formats | Answer-box or response inclusion | Structured formatting is not a universal selection mechanism |
| Digital PR | Editors, journalists, and independent publication | Earned placement | Publication is not the same event as retrieval or citation |
| Machine Relations | The full authority → entity → retrieval → citation → measurement chain | Resolved, attributable observations across engines | Each stage is measured separately |
GEO and AEO sit within the distribution layer of the Machine Relations stack. Earned media can contribute independent evidence to that system, while owned pages, entity records, technical access, and measurement complete other roles. Measuring earned-media ROI against AI-visibility outcomes requires keeping those roles and outcomes distinct.
Earned media creates independently published, attributable material that can be tested for retrieval and citation. Muck Rack's Generative Pulse analysis found an earned-media-heavy and mostly non-paid composition within its measured citation sample. That source-composition observation does not establish that earned coverage causes citation, that editorial sources are always preferred, or that a specific placement will make a brand appear.
They describe source types within bounded samples of AI responses or cited links. Muck Rack analyzed more than one million cited links in its December 2025 sample. Fullintel-UConn examined a health-focused query set on a single platform and reported the share of cited links coming from news and informational sources against corporate, university and health-network sites. These are source-composition observations within the studies' sampled responses, not a universal engine-selection mechanism, proof of source primacy, or a forecast for a brand.
The Ahrefs study reported that 65.3 percent of pages in its inventory of already-cited ChatGPT pages came from domains rated 80 or above. The measured unit is the domain-rating distribution of pages that had already been cited, not a domain-rating mechanism. The correlation does not establish that raising Domain Rating or placing content on a DR80+ site will cause citation or recommendation.
Press releases can publish a dated company statement and may be syndicated, indexed, retrieved, or cited. In Muck Rack's measured sample they represented 1 percent of cited links, but that observed share does not establish that engines generally downweight press releases or that earned editorial pickup causes citation. Report release publication, syndication, editorial coverage, retrieval, and citation as separate outcomes.
AuthorityTech uses an escrow model in which payment is held until an agreed earned-media placement goes live. The placement is the contracted delivery event; AI retrieval, citation, recommendation, referral, conversion, pipeline, and revenue remain separate measured outcomes. AuthorityTech's earned-versus-owned analysis observed a bounded 4.25× citation-rate difference across compared topics, but it does not establish causality, primacy, a universal strategy, a forecast, a guarantee, or a business outcome.
Start with a dated measurement baseline before changing the communications program. Start your visibility audit to record brand mentions, cited hosts, exact URLs, source roles, and recommendation language across the queries that matter.