When a buyer in 2026 wants to know who the experts are in a category, they no longer start with the editor’s choice. They ask ChatGPT. They ask Claude. They ask Perplexity. They see a Google AI Overview at the top of their search. The names that get repeated inside those answers become the experts in the buyer’s mind — regardless of which publication originally established them.
This is the new top of the authority stack: AI citation. And it changes how earned media works — not whether it works.
The Data
Across multiple citation-tracking studies, retrieval analyses, and AI visibility audits conducted through 2025 and 2026 — including independent research from Similarweb, SEMrush, Profound, Peec AI, Goodie, SE Ranking, Ahrefs, Evertune, and Passionfruit, and 5W’s Q1 2026 Citation Source Audit synthesizing the body of work — a consistent pattern has emerged across hundreds of millions of real AI citations and prompts.
Two findings reset the conversation.
1. The most-cited sources on the major AI engines are not the traditional Tier 1 outlets. On Gemini, the top five sources the model cites are Reddit, YouTube, Wikipedia, Medium, and Forbes. On ChatGPT, Wikipedia and Reddit alone account for over 25% of all citations — more than any traditional media category combined.
2. The Wall Street Journal, The New York Times, Bloomberg, and the Financial Times do not appear in the top 20 most-cited domains on ChatGPT. Three structural reasons: paywalls block clean extraction, licensing disputes with the AI companies have reduced indexing, and long-form narrative produces less extractable content than tighter, structured pieces.
The instinct after reading those two findings is to conclude that legacy media has lost. The conclusion is wrong. Legacy media did not lose. It moved.
The Authority Flywheel
The system that produces AI citation share is not a stack. It is a flywheel. Each layer feeds the next.
Earned Media → Wikipedia → AI Citation → Buyer Trust → More Earned Media
A Wall Street Journal feature generates impressions for ninety days and produces a citation-eligible source. Wikipedia editors quote that source, and the brand’s Wikipedia entry strengthens. ChatGPT, Claude, Gemini, and Perplexity retrieve Wikipedia and the surrounding citation graph, so the brand appears inside the answers buyers ask. Buyer trust compounds. Reporters at the next round of outlets see the brand cited inside AI answers and use it as a reference, which generates the next round of earned media.
The brands gaining AI Citation Share are not running this flywheel by accident. They are operating it intentionally. The brands losing share are running one or two stages and expecting the whole flywheel to spin.
The Two-Layer Stack
Layer 1 — Earned Media (the upstream layer)
Tier 1 coverage, trade press, contributor pieces, podcast appearances, and conference visibility. This layer’s job is to generate the source material that the AI engines eventually cite. A Wall Street Journal article doesn’t get cited directly by ChatGPT very often. But the WSJ article gets quoted in Wikipedia. The WSJ article gets referenced in a Forbes contributor piece that does get cited. The WSJ article shapes how the journalists at Reuters and the editors at Wikipedia describe the company.
Earned media is upstream feedstock. It populates the credibility graph the AI engines pull from. Without it, the rest of the flywheel has nothing to spin on.
Layer 2 — AI Citation (the new top of the stack)
The platforms where buyer questions get answered now: ChatGPT, Claude, Perplexity, Gemini, Google AI Overviews. The sources those engines pull from at retrieval time: Wikipedia, Reddit, YouTube, LinkedIn, Medium, structured Q&A interview platforms like Authority Magazine, review sites like G2, Capterra, and Trustpilot, analyst platforms like Gartner, and a specific subset of Tier 1 and trade media that survives the extraction filter.
A program with strong Layer 1 and weak Layer 2 produces earned media that doesn’t compound. The pickups happen, the impressions count, and the article ages out in 90 days. Nothing repeats inside the AI answers because nothing got translated into the structures the engines retrieve from.
A program with strong Layer 2 and weak Layer 1 produces AI citations built on weak foundations — citations that flip the moment a competitor publishes better upstream source material.
The complete program operates both.
The Authority Builders
A brand’s AI Citation Share is built across a defined set of sources the engines pull from. Most communications programs underweight this list because it does not fit the traditional PR map.
- Reddit — top-cited source on every major engine. Substantive long-form discussion, structured subreddit communities, content licensing partnerships with OpenAI and Google.
- Wikipedia — most-cited single source in ChatGPT. The ground truth layer for factual answers.
- YouTube — strongest single correlate of AI visibility in published research. Transcripts persist indefinitely.
- LinkedIn — fastest-rising signal. Named-leader publishing is now a ranking factor.
- G2, Capterra, Trustpilot — review platforms that drive recommendation citations. Brands present on the three majors see roughly 3x the citation share of brands absent from them.
- Gartner, Forrester — analyst sources that carry disproportionate weight on B2B retrieval.
- Medium — top-five source on Gemini. Hosts the structured Q&A interview networks the engines extract from cleanly.
- Tier 1 contributor networks — Forbes, Inc., Entrepreneur. The columns survive the extraction filter when the main news pages do not.
Each one is a different surface. Each one has different rules. A communications program that does not operate across all of them is not operating in the AI citation layer at all.
What Actually Happens Inside the Engines
ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews each have distinct citation patterns. There is no single “AI SEO” strategy that maps across platforms.
- ChatGPT is the most Wikipedia-heavy. Strong Wikipedia content disproportionately moves ChatGPT visibility.
- Google AI Mode favors Google-owned properties — YouTube, Google.com, Fandom.
- Gemini integrates Google search results directly. Strong traditional SEO converts to Gemini visibility better than to any other engine.
- Perplexity skews toward research-credible sources: NIH, G2, Gartner, structured B2B data. Most footnote-explicit of the engines.
- Google AI Overviews features YouTube, Reddit, Forbes, LinkedIn, and Wikipedia.
The pattern that holds across every platform: structured, third-party, repeatedly mentioned content beats prestige. Earned media has to be optimized not just for the reader and the editor, but for the extraction logic of the retrieval systems that now sit downstream of every article published.
The instinct in the industry has been to treat AI citation as a threat to traditional PR. The data shows the opposite. Earned media in citation-eligible publications is now the leverage behind AI visibility — and the brands with the most active earned-media programs are pulling away from the brands without them.
A single high-quality earned hit now:
- Gets read by humans for roughly 90 days, the traditional impression curve.
- Gets quoted in Wikipedia by editors who treat the source as citation-eligible.
- Gets referenced in Forbes, Inc., and contributor-network pieces that survive the extraction filter.
- Gets repeated inside answer-engine retrievals for years afterward.
- Compounds on every subsequent earned hit in the same credibility graph.
A press release without an earned-media foundation generates none of this. The release dies. The brand learns nothing. The AI engines never hear about the company.
This is why earned media has become more valuable, not less. The single Tier 1 placement that used to generate a 90-day visibility curve now generates a six-year compounding curve — if and only if the rest of the flywheel is built to capture it.
The Six Levers
The brands gaining AI Citation Share are operating six levers simultaneously.
1. Earned media in citation-eligible publications. Forbes, Reuters, trade press, Tier 1 contributor networks. Coverage that survives the extraction filter and that Wikipedia editors treat as citable.
2. Wikipedia as infrastructure. Wikipedia is the single most-cited domain in ChatGPT. A weak or missing Wikipedia page is the single biggest cap on AI Citation Share. The path is not direct editing — it is earning citation-eligible coverage that other editors then use to build the page.
3. Distributed authority across the platforms the engines pull from. LinkedIn long-form publishing under named leaders. Structured Q&A interviews on platforms like Authority Magazine. YouTube category mentions that survive transcript indexing. Review platforms (G2, Capterra, Trustpilot) for verticals where they apply. Analyst presence on Gartner and Forrester for B2B.
4. Extractable content on owned properties. Case studies with specific numbers, named clients, structured outcomes. FAQs that answer one question per page in clear, extractable language. Deep vertical reference pages that own a single topic decisively. Schema markup on every page that matters.
5. Repetition across many sources. Distribution across many citation-eligible domains beats concentration in a single one. The brand whose name appears consistently across thirty mentions in thirty different surfaces beats the brand with ten mentions in three.
6. Entity density. Brands. People. Products. Categories. Named consistently across every surface. AI engines build their answers around named entities — and they reward the consistent, repeated naming of the entity-relationships that define a category. A communications program that buries the brand name, the product name, the founder name, and the category name in passive prose generates fewer citations than a program that names them clearly and repeatedly across every surface it touches. Entity density is the most underweighted concept in GEO. It is also the simplest to fix.
None of these levers replaces earned media. Each one extends it.
What AI Citation Is Not
The discipline of AI Communications gets misunderstood as quickly as it gets named. Four definitional notes prevent confusion.
AI citation is not keyword stuffing. The retrieval engines have moved past the lexical-match logic of 2010-era SEO. Filling a page with target phrases generates no citation lift and frequently triggers extraction penalties.
AI citation is not prompt hacking. Embedding hidden instructions inside content, manipulating LLM responses through adversarial text, or attempting to inject behavior into the engines is a different discipline — adversarial machine learning, not communications — and it does not scale, does not survive model updates, and exposes the brand to reputational and platform-level risk.
AI citation is not AI spam. Generating low-quality content at volume in the hope of getting indexed produces the opposite outcome. The engines filter aggressively for source quality, originality, and extraction signal. Volume without substance triggers downranking.
AI citation is not fake reviews, fake citations, or fake authorship. Every retrieval engine treats source credibility as a primary input. Manufactured Reddit threads, paid Wikipedia edits, fabricated review profiles, and ghost-authored Medium posts get detected, filtered, and frequently penalized at the platform level.
AI Communications is the same discipline as traditional public relations, applied to the new retrieval layer. Real earned media. Real authority. Real distribution. Measured against the engines that now answer the question. The shortcuts do not work — and the firms that try them lose ground.
What to Do This Quarter
Three moves separate the firms operating the new stack from the firms still measuring against the old one.
Audit your existing earned-media program against the citation graph. A pickup count is not the metric. The metric is which of your earned hits actually populates the sources AI engines pull from. A WSJ feature that’s never cited in your Wikipedia page is leverage you paid for and never collected.
Build the Layer 2 infrastructure your category requires. Wikipedia, structured Q&A interview presence, LinkedIn long-form publishing on a named-leader cadence, YouTube transcript visibility, review platform completeness, and analyst presence — calibrated to the verticals where your buyers ask AI questions.
Move from annual measurement to quarterly. AI citation patterns shift on a multi-week timescale. Published research documents a single two-week window in September 2025 when Reddit’s share of ChatGPT citations collapsed from approximately 60% to 10%. Annual reporting against AI visibility is measuring against patterns that no longer exist.
The Bottom Line
The authority stack did not replace public relations. It expanded it.
Brands that understand only earned media will become less visible inside the answer engines that now drive buyer research. Brands that understand only AI citation will lack the upstream credibility that makes those citations stick. The winners will build both.
