We looked for a human at all 439 domains only one AI engine cites. 9.3% publish a named person's email.

Method, stated before the numbers

The population is the 439 domains our 2026-07-28 citation cut found cited by exactly one of ChatGPT, Perplexity, Gemini or Claude across 144 answers to 12 pinned buyer questions — the long tail that makes up 81.6% of everything the engines cite on these questions. For each domain we issued an HTTPS GET against 8 conventional paths (`/`, `/contact`, `/contact-us`, `/about`, `/about-us`, `/team`, `/privacy`, `/privacy-policy`), extracted every email address in the raw HTML, kept only addresses on the outlet's own registrable domain, and classified each local part against an 86-term role-alias list (info@, support@, press@, privacy@ and the like), followed by a manual second pass that demoted role-shaped survivors the list missed (dpo@, customersuccess@, founders@) and one literal `john.doe@` documentation placeholder. One sweep, one day, every rejection recorded.

What 439 cited domains publish

Read the bottom row first, because every piece of “get cited by the sources AI already cites” advice quietly assumes it is large: of the 439 domains that make up the single-engine citation tail, 41 publish an email address that belongs to a person with a name. The other 90.7% offer a contact form, a role alias, or nothing.

The most-cited tail domains are the least reachable kind

The single-engine tail is not 439 small blogs. Its most-cited members are hubspot.com (11 answers — role aliases only), techradar.com (9 — a masthead), linkedin.com and medium.com and arxiv.org (6, 5 and 5 — no published address, because they are platforms, and the thing the engine actually cited is a page some third party posted there). A business cannot pitch arxiv.org for coverage; the citation belongs to whoever happened to publish there. A meaningful share of the AI citation layer is not an outlet you can build a relationship with — it is infrastructure other people publish on.

At the other end sits one outlier worth naming: seerinteractive.com publishes 418 individually named staff addresses on its own pages — more named humans on one domain than the other 438 domains publish combined.

Who the 41 are

Overwhelmingly: founders of small marketing, SEO and AI-visibility consultancies who publish their own first-name address — dan@, mantas@, jolyn@, graham@, chris.wright@ — plus a handful of large-outlet mastheads (TechRadar, Tom's Guide, Entrepreneur). The pattern inverts the usual PR logic: the bigger the cited domain, the less likely you can reach a named person; the domains where a human answers the mail are one-person shops whose citation you could plausibly earn yourself instead of borrowing.

Limits

• 8 paths, raw HTML only. Addresses rendered by JavaScript, obfuscated, or on unconventional paths are missed, so every published-address rate here is a floor, not a ceiling.
• “Unreachable” includes bot-blocking. A domain behind aggressive bot protection counts as unreachable in this sweep even though a browser would load it; 56 is an upper bound on genuinely dead domains.
• The role-alias line is a heuristic. An 86-term list plus one manual pass; a first-name address can still be a shared inbox, and a role alias is sometimes read by exactly one person.
• One sweep of one corpus. The 439 domains come from a single 144-answer run on 2026-07-28; the tail churns run to run, and this is a one-time measurement, not a tracked series.
• We have a commercial interest. We sell AI-visibility audits, and we contact domains from this population ourselves — identity-transparent, from this company, at most a handful a day.

The class-by-class table above and the per-domain classification behind it are a first-party cut we are happy to share with anyone who wants to check it. For where these 439 sit in the wider citation economy, read the four-engine overlap cut; to see which engines cite anything about your own business, you can run the same measurement on your business in 60 seconds.