Somewhere in the last eighteen months, the first question a potential client, recruit, or investor asks about an unfamiliar executive quietly changed. It used to be a Google search. Increasingly, it’s a prompt — “who’s the leading voice on supply chain resilience,” “who should I talk to about brand strategy in consumer products” — typed into ChatGPT, Claude, or Gemini. The system doesn’t return ten blue links for a human to sort through. It returns an answer, with two or three names attached, and moves on. Whichever name isn’t in that answer effectively doesn’t exist for the person asking.
That makes the mechanism behind the answer worth understanding in some detail, because it turns out to be unusually well documented. Three separate research efforts published in 2026 — from a media-intelligence company, an AI-visibility startup, and a marketing-analytics firm — independently measured what these systems actually cite, at a scale nobody in executive branding was tracking closely until this year. They agree with each other almost exactly, and the agreement is not flattering to the instinct most executives have to just publish more content on their own site.
What 25 million citations actually point to
Muck Rack’s May 2026 edition of its ongoing “What Is AI Reading?” study is the largest dataset of the three — more than 25 million links cited by ChatGPT, Claude, and Gemini, tracked across 17 industries. The finding, consistent with the two prior editions going back to July 2025, is that earned media — independent journalism and third-party editorial coverage — accounts for 84% of everything these systems cite. Paid and advertorial content, the category that includes sponsored posts and press releases dressed up as articles, accounts for 0.3%. Journalism alone, not counting trade publications or industry outlets, makes up 27% of citations by itself.
Read plainly, that means when an AI system is asked to explain who matters in a given field, it is reconstructing the answer almost entirely from coverage nobody paid for and nobody at the company wrote. It is functioning less like a search engine and more like an unusually well-read editor, deciding whose byline to trust based on who else already vouched for them in print.
Why an executive’s own site barely shows up
The instinct this creates a problem for is the most common one in executive content strategy: publish more, on channels you control. Research from Ranqo, an AI-visibility analytics firm, measured exactly how far that gets an executive. Across the citations it tracked, only 2.9% pointed back to a brand’s own domain — not a rounding error, a near-absence. In a distribution pilot Ranqo ran in December 2025, the same piece of content saw its citation rate move from 7.7% to 34%, a 325% increase, the moment it ran through third-party outlets instead of staying confined to the brand’s own channels.
This is the same dynamic this series has traced through earned press coverage and Wikipedia notability: the version of a claim that a self-interested party makes about themselves carries almost no evidentiary weight, no matter where it’s published. The version an independent outlet chooses to write, unprompted, is what every downstream system — a nominating committee, a Wikipedia editor, now an AI model — treats as real signal. AI search didn’t invent that hierarchy. It just automated checking for it, at a scale no human editor could match.
Citation compounds on top of authority that already exists
The third dataset, from the AI-optimization firm AirOps, gets at a subtler mechanic: how AI models choose among the sources they’ve already decided are legitimate. Analyzing 548,534 pages ChatGPT retrieved across 15,000 real prompts, AirOps found that the model reads far more than it ever cites — only 15% of retrieved pages made it into a visible answer. What separated the 15% from the other 85% tracked closely with existing search authority: pages already ranking #1 on Google were cited 3.5 times more often than anything ranking outside the top 20. Query type mattered too — practical “how-to” and product-discovery questions saw citation rates around 17–18%, while comparison and validation queries, the ones closest to “who is the real authority here,” saw the lowest rates of all, in the 11–13% range, suggesting the model is being most conservative exactly where an executive’s reputation is on the line.
The takeaway compounds rather than replaces what drives visibility everywhere else in an executive’s public record: a name that already has real search authority, built from durable, independently earned coverage, is the name an AI system is statistically most likely to trust when it’s deciding who to name out loud. There’s no separate “AI visibility” track to build. There’s the same track — get written about by people with no reason to flatter you — that this series keeps returning to, now with a new, unusually literal-minded reader checking the receipts before it repeats your name.
The answer machine doesn’t take submissions
None of the three studies found a shortcut, and the absence of one is the finding. An executive cannot submit their own bio to ChatGPT the way they might submit a nomination form for an award. They cannot buy a citation the way a “guaranteed Wikipedia page” vendor promises to sell one. What they can do is the slower, harder version of the same work every channel in this series keeps circling back to: get profiled, get quoted, get covered by outlets with their own earned authority — and let the machine read that record the way it was built to be read, by someone else, about them.
The executives who show up when someone asks an AI system who the authority is on their subject are, almost without exception, the ones who were already the answer to that question before anyone thought to ask a machine. The citation isn’t the achievement. It’s the receipt.