Earned Media as an AI Credibility Signal: What Each Engine Documents
No AI engine publishes a weight for earned media. What each documents instead, with dates: Google, Bing, Perplexity, ChatGPT and Claude as of October 3, 2026.
None of the five engine operators we checked documents a ranking weight for "earned media" as a category. What each documents is a different layer where credibility is judged: Google's quality and spam systems, Bing's evidence-and-structure guidance, Perplexity's domain-level source labels, OpenAI's unspecified multi-factor ranking, and Anthropic's operator-controlled domain lists. This page records what each says, with dates, as read on October 3, 2026.
Studies that count what engines cite are a separate question. Muck Rack's May 2026 "What Is AI Reading?" measured the composition of more than 25 million cited links; its May 7, 2026 release reports earned media at 84% of citations. The category-level reading of earned versus owned citation rates is kept in the Machine Relations canon: Earned vs. Owned AI Citation Rates, published July 6, 2026. This page asks the mechanism question those counts leave open: what do the engines say they do with a credibility signal?
What each engine documents
| Engine | Documented credibility layer | Not documented | Primary source, with date |
|---|---|---|---|
| Google (AI features in Search) | Generative features are "rooted in our core Search ranking and quality systems"; spam systems filter inputs | Any weight for press mentions or earned coverage | AI optimization guide, last updated July 10, 2026 |
| Bing / Copilot | Content "evaluated for authority and relevance"; publishers are told to supply evidence, structure and freshness | A way to convert mentions into selection; the AI Performance report states citation counts are not a ranking or authority measure | Bing Webmaster Blog, February 10, 2026; Microsoft Advertising, October 8, 2025 |
| Perplexity | Domain-level labels (Government, Academic, Trusted) assigned on editorial-practice criteria | Any claim that a label raises retrieval or ranking | Understanding source labels, last modified September 9, 2026 |
| ChatGPT search | Results ranked on "multiple factors intended to help users find relevant, reliable information"; placement not guaranteed | The factors; any earned-media weighting | Searching the web with ChatGPT, as read October 3, 2026 (no revision date shown) |
| Claude | Web search with citations; operators on the API can allow or block domains | Any credibility weighting beyond that | Claude web search help and web search tool docs, as read October 3, 2026 |
The absence of a documented weight is not proof that no weighting exists. It is a limit on what can be stated from first-party sources. Anything stronger about why a particular outlet is cited is inference from observed citations, not from engine documentation.
What Google says about mentions
Google's AI optimization guide, introduced in a May 15, 2026 Search Central post and last updated July 10, 2026, is the most explicit of the five on this question. It says Google's generative AI features "are rooted in our core Search ranking and quality systems," and that "seeking inauthentic 'mentions' across the web isn't as helpful as it might seem," because the core ranking systems focus on high-quality content while other systems block spam, and the generative features depend on both.
Read narrowly, that sentence does not say earned coverage is discounted. It says mentions pursued for their own sake are not a shortcut, and that the same ranking and spam machinery that governs classic results governs the generative layer. Google's spam policies, last updated August 28, 2026, list attempts to manipulate generative AI responses in Google Search as spam. Our earlier look at the policy side is in Google AI search spam policy and generative responses.
What Bing and Microsoft say about citations
Bing's AI Performance report, in public preview since February 10, 2026, gives publishers counts of how often their pages are cited. The announcement is explicit that a citation count is not a ranking, authority or page-importance score. Microsoft Advertising's October 8, 2025 guidance says "there's no secret strategy for being selected by AI systems" and frames selection as content evaluated for authority and relevance. The practical advice in both is evidence, structure and freshness. The metric view is covered in Bing Webmaster citation share as an AI visibility metric.
Where earned versus paid shows up in documentation
The nearest first-party statement of an earned-versus-paid boundary is Perplexity's source labels. The page describes labels applied at the domain level, based on editorial practices such as corrections and bylines and on news being kept separate from advertising and opinion. It states that payments and partnerships do not affect labels, that a label is not a claim of accuracy or ranking, and that an unlabeled domain is not a negative judgment. Muck Rack's finding that paid and advertorial content makes up 0.3% of cited links is consistent with that separation, but the two are different kinds of evidence: one is an observed distribution, the other a documented labeling rule. Neither shows that an engine ranks on the earned or paid status of a piece.
OpenAI and Anthropic document access and control, not credibility. OpenAI's page names OAI-SearchBot access as the eligibility step; Anthropic's documentation covers citations and domain lists. A buyer reading only those two sources cannot tell whether earned coverage helps.
What changed, dated
- February 10, 2026: Bing publishes AI Performance in public preview and states citation counts are not authority.
- May 15, 2026: Google introduces its generative AI optimization guide, tying generative features to core ranking, quality and spam systems.
- July 10, 2026: Google's guide last updated.
- August 28, 2026: Google's spam policies last updated; they name manipulation of generative AI responses.
- September 9, 2026: Perplexity's source-labels page last modified.
Engine-by-engine records of what is actually cited, rather than what is documented, are in the change logs: ChatGPT, Gemini, Perplexity and Claude. The per-engine volatility behind them is in Per-engine AI citation churn over 89 days. There is no Copilot change log yet.
What this means for the funding decision
A marketing or communications lead deciding whether to fund earned media as an AI-visibility lever should separate three claims.
- Earned coverage correlates with what engines cite. Supported by observed counts such as Muck Rack's and the Machine Relations canon.
- Engines weight earned coverage as a category. Not documented by any of the five operators.
- Mentions pursued for the purpose of AI visibility help. Google says they may not, and its spam policies cover manipulation of generative responses.
The documented levers are eligibility (crawler access and indexing), content that is evidenced, structured and fresh, and publication on domains whose editorial practices pass each engine's own checks. Earned coverage can be one way to be present on such domains. It is not a documented lever on its own, and the evidence for it is observational.
FAQ
Does any engine say it prefers earned media? Not in the first-party documentation we read. Google, Bing, Perplexity, OpenAI and Anthropic each describe credibility-related systems or controls, but none names earned media as a weighted category.
Does the 84% figure show engines weight earned media? It shows that earned media makes up most of the links Muck Rack counted in cited sources. A composition count does not identify why those links were selected. See Muck Rack's release for its method.
Are mentions pursued for AI visibility penalized? Google says seeking inauthentic mentions is not as helpful as it might seem and its spam policies cover attempts to manipulate generative responses. Other engines do not document a position.
Where do the engines differ most? Perplexity documents a domain-level labeling review; Google documents inheritance from core ranking and spam systems; OpenAI and Anthropic document access and controls. Bing documents a citation report and says it is not an authority score.
What would change this reading? A first-party statement from any operator that earned or editorial sources receive distinct weighting in selection, or a change in the sources above. Each is dated in the table so a later revision can be compared.