The distribution is not the evidence
When a company issues a press release and fifty publications pick it up, something real has happened. The release has circulated. The claim has propagated. The company’s name appears across a significant portion of the web’s active publishing surface. For traditional SEO purposes, that distribution matters: it produces links, mentions, and indexed content across authoritative domains.
For AI citation purposes, it produces something different: the appearance of independent corroboration where none exists. Fifty publications citing the same press release are not fifty witnesses. They are one document, fifty times. The underlying evidential count is one.
What AI systems actually evaluate
An AI system building an answer about a company encounters not the press release itself but its downstream propagation. It sees the trade publication that reprinted the key paragraph. The news aggregator that summarized the announcement. The industry newsletter that cited the trade publication. The analyst note that cited the newsletter. By the time the claim reaches the AI’s training data or live retrieval window, the original press release may be invisible — replaced entirely by its derivatives.
The derivatives look, from the outside, like independent sources. They carry different domain names, different publication dates, different authors, different surrounding content. A citation tracker counts each one. What it does not count is how many independent parties actually investigated the underlying claim.
The press release is a designed propagation instrument
This is not a criticism of press releases. A press release is an efficient tool for distributing a claim to as many publishers as possible in the shortest time. It is designed to maximize pickup, standardize language, and reduce the editorial friction that would otherwise filter or reframe the underlying claim. It works exactly as intended.
The design also means that successful press release distribution produces a specific pattern: high publisher count, low evidence independence, rapid propagation across domains that would otherwise represent diverse, independent sources. A company that issues a single press release and earns fifty articles of coverage has not produced fifty independent corroborations of its claims. It has produced one assertion in fifty containers.
The question is not how many places said it. The question is how many independent parties investigated it.
Infrastructure signals see through the distribution layer
Domain trust infrastructure does not count publications. It observes the accumulated record that independent parties — not press-release recipients — have built around a domain over time. A domain that has been consistently present in authoritative network neighborhoods, referenced in non-promotional contexts, and observed across multiple independent collection events carries a different signal than a domain that appears primarily in press-release distribution patterns.
This is why infrastructure signals and content signals answer different questions. A content signal — how many articles mention this company — will reflect the success of a press release campaign. An infrastructure signal — how does the network independently position this domain — will not. The two measurements can diverge sharply. A company that has issued many successful press releases and earned wide coverage may score poorly on infrastructure signals if the coverage never translated into the kind of organic, independent reference that builds domain trust over time.
What this means for AI citation
An AI system that retrieves claims from multiple publications about the same announcement is not retrieving multiple independent witnesses. It is retrieving the same press release in different fonts. The practical consequence is that companies with high press release velocity and low infrastructure trust may appear well-represented in AI answers for a period, then find that representation unstable — because the underlying signal the AI is actually evaluating is not the press release count but something older and less manufactured.
The implication runs in both directions. A company with deep infrastructure trust and few press releases may be more stably represented in AI answers than a company with many press releases and shallow infrastructure trust. And a company that understands this distinction can stop measuring press release pickup as an AI visibility metric and start measuring the thing the AI is actually using: the accumulated, independent, non-promotional signal that the domain has been building, or failing to build, for years.