Earned Media as AI Visibility Infrastructure

Summary

If your AI visibility program starts and ends on your website, you are optimizing the minority path.

Muck Rack's May 2026 Generative Pulse report — analyzing 25 million+ links across ChatGPT, Claude, and Gemini — found earned media drives 84% of AI citations. Paid/advertorial content accounts for 0.3%. Press release wire syndication: 0.04%. Journalism alone contributes 27%. That pattern has held between 82–89% across three measurement windows since July 2025.

Separately, Victorious's Q2 2026 report found AI accurately recognizes 96% of tested brands when asked directly — yet 89% never appear in category research answers. Of 49,391 citations in that study, 99.99% pointed to third-party sites rather than the brand's own domain.

This white paper argues earned media is no longer a brand-awareness side channel. It is AI visibility infrastructure — the retrieval layer answer engines trust when buyers ask category questions.

Key takeaways:

• Owned content builds entity clarity; third-party content wins most citations and mentions.
• Third-party mention volume correlates with AI mention rates (Victorious: under 2K mentions → 3% mention rate; 30K+ → 64%).
• Listicles and "best of" formats punch above weight (Evertune: 63% of LLM citations point to listicles in studied sets; content-level listicle share often ~35%+).
• Reddit's influence 8×es from TOFU to BOFU on Google AI Overviews for SaaS (EMGI).
• Claude barely cites social UGC — prestige editorial is the Claude path.
• PR, SEO, community, and AEO must share one prompt-driven scoreboard.

The evidence stack

1. Citation majority is earned

The implication is not subtle: your website is not the primary input to AI visibility. Third-party editorial coverage is.

Independent studies converge on the same range:

Article deep dive: 84% of AI Citations Come From Earned Media.

2. Recognition ≠ recommendation

Victorious (175 brands, 8 platforms):

• 96% accurately described on direct brand prompts
• 89% never appeared in category answers
• Referring domains r = 0.49 and third-party mentions r = 0.45 with AI mention rate
• Knowledge Graph presence: no stable relationship alone

Mention rate by indexed third-party mentions:

~20,000 mentions ≈ 50% probability of category mention in that dataset.

Article: AI Recognizes Brands but Rarely Mentions Them.

3. Citation without naming (ghost citations)

Semrush / Kevin Indig: 61.7% of appearances were ghost citations — URL cited, brand unnamed. Citation rate roughly double mention rate. ChatGPT is citation-heavy; Gemini is mention-heavy.

Article: Ghost Citations.

4. Domain authority still gates retrieval

Indexably (18,129 AI-cited pages): domain factors ~77% of predictive importance; page optimization lifts mainly the top authority quartile. Backlink diversity (referring subnets) beats raw link volume.

Article: Page Optimization Only Helps High-Authority Domains.

5. Community peaks at decision time

EMGI (1,486 SaaS queries): Google AI Overviews cite Reddit on 2.5% of TOFU vs 20.1% of BOFU queries; Reddit is on 94.1% of BOFU SERPs. Siege Media: Reddit in ~62% of BOFU LLM responses in their sample.

Article: Reddit AI Citations Concentrate at BOFU.

What "earned" means for AI systems

Muck Rack's earned bucket includes journalism, academic/government sources, encyclopedias, third-party editorial, reviews, and trusted community content — not brand-owned pages, paid placements, or advertorial.

AI engines prefer these sources because they reduce commercial bias in synthesis. The model can verify claims against independent pages. Owned marketing copy is discounted; wire PR is nearly invisible (0.04%).

Implication: pitching a journalist, earning a G2 category listing, or being named in a peer Reddit thread is not "awareness." It is adding a retrievable evidence node to the graphs engines use at answer time.

Platform-specific earned strategies

Vertical citation pools also differ (Victorious): legal concentrates on Vault/Chambers-class directories; healthcare on NIH/Mayo; SaaS spreads across 10,000+ domains including G2, Reddit, LinkedIn, Gartner.

One PR blast does not equal multi-engine earned strategy.

The earned media operating model

1. Inventory the footprint

For each priority brand:

• Indexed third-party mention estimate (excluding owned domain)
• Presence on review platforms (G2, Capterra, TrustRadius, industry-specific)
• Wikipedia / Wikidata eligibility and accuracy
• Top listicles and "best of" pages that already rank or get AI-cited in your category
• Specialist subreddits and YouTube channels that appear in BOFU answers
• Prestige / trade editorial hits in the last 12–24 months

Score against the Victorious thresholds. If you are under ~2,000 meaningful mentions, category mention probability is near floor.

2. Prioritize formats engines absorb

Highest leverage for many commercial categories

1. Ranked listicles / roundups (Evertune listicle dominance)
2. Independent journalism and original research that journalists can cite
3. Review platform presence with named brand entities
4. Buyer-shaped community threads (question titles, peer answers)
5. YouTube explainers/demos (for Google/Perplexity paths)

Low leverage for AI citation

• Wire syndication as primary tactic
• Paid advertorial expecting citation lift
• Brand-only blog publishing without distribution

3. Align pitches to buyer prompts — not brand slogans

Build a shared prompt library with AEO. For each gap ("competitors named, we are not"):

• Which third-party URL is currently cited?
• Can we earn inclusion/update on that URL?
• Or create research that spawns new citable coverage?

PR briefs should include the exact prompts you want to win — not only narrative themes.

4. Coordinate the four teams

Weekly stand-up: one prompt gap board, owners per gap, no vanity "mentions" without answer-layer validation.

5. Measure earned as infrastructure KPIs

Framework: Measuring AI Visibility.

Original research as a force multiplier

Journalism's 27% citation share rewards newsworthy data. Brands that publish original surveys, benchmarks, and category datasets create:

1. Direct owned citations (minority path)
2. Downstream editorial citations (majority path)
3. Mentions that raise third-party inventory

One strong study can generate dozens of earned nodes. Treat research as AEO capital expenditure.

What owned content is still for

Owned pages are not useless. They:

• Establish entity facts models must get right
• Feed retrieval when engines do cite owned domains
• Support Bing/Google/Brave indexing
• Convert the minority of users who click through

Indexably and Martinez's GEO survey both warn against body-only rewrite theater. Pair owned excellence with earned distribution — sequenced by authority tier.

Listicles as earned infrastructure (not content vanity)

Evertune's analysis of nearly 400 million citations across 25,000 URLs found 63% pointed to listicle-format pages. Within content-type studies, listicles often account for ~35%+ of cited content pages — far above their share of the open web.

For category prompts ("best X for Y," "top tools for…"), the retrieval path frequently runs through a third-party roundup before it ever touches your homepage. That means:

1. Inclusion on existing high-citation listicles is a first-order AEO task
2. Publishing another owned listicle without distribution is usually a second-order task
3. Updating stale listicles that already get cited beats inventing new ones nobody links

Treat listicle outreach like link building with an answer-layer KPI: did your brand's mention rate rise on the prompts that listicle serves?

Detail: AI Search Loves Listicles.

The recognition trap (and how PR teams fall into it)

Victorious's finding — 96% recognition, 89% never mentioned in category answers — maps onto a familiar PR failure mode:

• Leadership asks: "Does ChatGPT know who we are?"
• Team demos a brand prompt: accurate description
• Leadership concludes: "AI visibility is fine"
• Buyers ask category prompts: competitors occupy the shortlist

Brand prompts measure entity clarity. Category prompts measure recommendation. Earned media primarily moves the second. Report both — never substitute one for the other.

Vertical playbooks (abbreviated)

Copying a SaaS Reddit strategy into healthcare is how programs burn budget without moving Claude or Gemini citations.

90-day earned infrastructure plan

Days 1–30 — Baseline

• Mention inventory + competitive SOV by engine
• Map top cited third-party URLs for BOFU prompts
• Identify 10 listicle/review targets and 5 research angles

Days 31–60 — Placement

• Pitch / contribute to listicles and trade coverage
• Fix review profiles; solicit authentic reviews
• Seed or earn presence in specialist community threads (non-astroturf)
• Ship one original data asset

Days 61–90 — Validate

• Re-measure category mention/SOV
• Attribute which new URLs appear in answers
• Double down on surfaces that moved metrics; cut vanity placements

How Obsurfable fits

Obsurfable shows whether earned placements actually appear in AI answers for your buyer prompts — including competitor substitution when a Forbes piece or Reddit thread names someone else. The Visibility Director connects prompt gaps to content and distribution actions.

Companion papers: Multi-Engine AEO Operating System · Measuring AI Visibility · AEO & GEO Report

FAQ

Can we buy AI citations?

Paid/advertorial share is ~0.3%. Buying ads does not buy citation inclusion. Earn independent coverage.

Should we stop blogging?

No. Stop treating the blog as the primary citation strategy. Use it for depth and conversion; use earned for recommendation.

Does Wikipedia solve everything?

It helps ChatGPT entity grounding when you qualify. It does not replace review platforms, listicles, or BOFU community presence.

What about Claude specifically?

Prioritize prestige and institutional coverage. Reddit/YouTube programs will barely move Claude.

Bottom line

Answer engines do not primarily recommend the brands with the best homepages. They recommend brands the rest of the web repeatedly names in trusted contexts. Earned media — journalism, listicles, reviews, specialist communities, prestige editorial — is the infrastructure layer of AI visibility. Build it deliberately, measure it on category prompts, and stop confusing owned publishing with being chosen.