AI search news ·
The AI-citation outreach playbook has almost nobody to write to: 7 of 99 multi-engine-cited domains publish a person's email address
Seven of the 99 domains that two or more of ChatGPT, Perplexity, Gemini and Claude cite on small-business AI-visibility questions publish an email address belonging to a person. Sixty-eight publish only a role alias — support@, sales@, hello@. Twenty-three publish no address at all. One refused our requests entirely. That is a 7.1% addressable surface for the tactic every GEO playbook recommends, measured on 2026-07-30 across 792 requests that returned 383 pages, and we ran it because we needed the answer for ourselves.
Primary source: AskedAbout — four-engine citation panel (2026-07-28) and contact-address sweep (2026-07-30)
Why we measured this
The standard advice for getting recommended by AI engines has two halves. Be citable, and pitch the people the engines already cite. The second half assumes something nobody seems to have checked: that the domains an engine cites can be reached by a human writing an email. We have been running that tactic for five weeks, so we checked.
The population is the one we published on 2026-07-29 in Only 10 of 538 domains were cited by all four AI engines: twelve pinned buyer questions — “What are the best AI visibility tools for small businesses?”, “How do I improve my business's visibility in ChatGPT and AI search?” and ten more — run three times each against OpenAI, Perplexity, Google Gemini and Anthropic Claude on 2026-07-28. 144 answers, 131 carrying a machine-readable citation payload, 538 distinct domains. This sweep takes the 99 of those domains cited by two or more of the four engines, on the reasoning that a domain a single engine cites once is not worth an email either way.
Method, stated before the numbers
For each of the 99 domains we requested eight conventional contact paths over HTTPS — `/`, `/contact`, `/contact-us`, `/about`, `/about-us`, `/team`, `/privacy`, `/privacy-policy` — and extracted every email address appearing in the raw HTML, including addresses inside JSON-LD blocks. 792 fetches, all on 2026-07-30. We then discarded (a) every address not on the outlet's own registrable domain, which removes testimonial screenshots and embedded vendor addresses, and (b) placeholder and telemetry strings such as `you@company.com` and Sentry ingest keys. What survived was classified by its local part against a fixed 70-term role-alias list (`support`, `sales`, `hello`, `info`, `press`, `privacy`, `founder`, `careers`, and so on). Anything not on that list counts as person-shaped. The 792 requests returned 383 pages; the rest were 404s or refusals, and a domain counts as swept when at least one of its eight paths returned.
Three limits, because the number is only useful with them attached. This reads raw HTML only — an address rendered by JavaScript, hidden behind a contact form, or set as an image is invisible to it, so 7 of 99 is a floor, not a ceiling. We fetched each path once, with no login, no form submission and nothing behind a paywall. And one domain, americanbar.org, refused all eight requests; it is counted in its own row rather than folded into “no address”.
The result, by how many engines cite the domain
| Cited by… | Domains | Person-shaped address | Role alias only | No address in raw HTML | Refused our requests |
|---|---|---|---|---|---|
| All four engines | 10 | 2 | 5 | 3 | 0 |
| Exactly three engines | 28 | 0 | 26 | 2 | 0 |
| Exactly two engines | 61 | 5 | 37 | 18 | 1 |
| Total | 99 | 7 (7.1%) | 68 (68.7%) | 23 (23.2%) | 1 |
The three-engine tier is the striking row: 28 domains, not one person-shaped address between them. The mechanism is not mysterious once you look at what is in these tiers. Roughly nine in ten of the multi-engine-cited domains are SaaS tool sites, and a SaaS tool site publishes `support@`, `sales@` or nothing at all. The site shape that does publish a human address — the sole-principal consultancy, the one-person newsletter — is almost absent from the top of this population by construction.
The seven, one line each
A person-shaped local part is not the same thing as a reachable person, so each of the seven was then checked by hand for two things: is that person named somewhere on the same domain, and does the domain publish a geography? Both matter if you intend to send compliant mail rather than just any mail.
| Domain | Engines citing it | Address form | Person named on the same domain? | Geography published |
|---|---|---|---|---|
| aiclicks.io | 4 | first name | Yes — founder, in JSON-LD | None found |
| useomnia.com | 4 | first name | No — address appears only as a JSON-LD contact point | Madrid, Spain |
| pedowitzgroup.com | 2 | first name | Yes — founder and CEO, named in JSON-LD and beside the address | Milton, Georgia, US |
| juliareneeconsulting.com | 2 | first name | Yes — sole principal, invites email directly | United States |
| visiblie.com | 2 | two first names | Yes — both founders, in JSON-LD | Ghent, Belgium |
| triplewhale.com | 2 | first name + initial | No — that name appears nowhere in the rendered page | Columbus, US |
| foglift.io | 2 | single word | No | None found |
So of the 99: four domains publish an address you can tie to a named human on the same site. Two of the other three are name-shaped mailboxes with no person attached anywhere on the site — a `firstname_initial@` given as a press contact, and a single-word mailbox that could as easily be a product. Whether you treat those two as reachable people is a judgement call; we do not, because an address you cannot attribute is an address you cannot personalise, and a wrong guess spends the deliverability of every other message you send.
Our own number, which is worse than the addresses
We publish our own figures on this beat every week, so: across five weeks we have sent identity-transparent pitches to 25 named humans on and around this population. Sixteen opened them. None has ever replied — not a yes, not a no, not a “please stop”. Our own domain, askedabout.com, appears in zero of the 1,465 citation entries in the panel above. The address scarcity documented here is real, and it is still not our binding constraint. The reply rate is.
That ordering matters if you are budgeting effort. A 7.1% addressable surface makes the pitch tactic small. A 0-for-25 reply rate on the addresses you do find makes it slow as well, and the two compound: you cannot fix a response-rate problem by sweeping more domains, because the domains do not have addresses.
What we would do with this, and what we are doing
- Check the addressable surface before committing to an outreach programme. For this population it is 7 domains out of 99, and the tier where a citation is worth the most — three engines — returned zero. That is a planning number, not a detail.
- Do not substitute role aliases when the named addresses run out. Every bounce we have ever taken, on any campaign, was a role alias; none was a named person. The alias is where the outreach quietly turns into spam, and it costs the deliverability of everything else you send.
- Weight effort towards being citable rather than pitching the cited. 439 of the 538 domains in the panel were cited by exactly one engine, which means the citation set churns; the durable move is publishing the specific answer a question needs.
- Publish the method with the number. Ours is a raw-HTML sweep on one date, so it undercounts JavaScript-rendered addresses. Anyone can rerun it and get a different figure; they should be able to see why.
The full dataset behind this post — all 99 domains with the engines citing each, the addresses found, the classification and the hand-checked notes on the seven — exists as a single JSON file. If you want it to check our arithmetic or to run the sweep yourself, ask and we will send it; there is nothing to sign up for.
See your number
A free 60-second check shows what AI says about you.
Running this for clients? The $249 agency 5-pack audits five businesses, white-labeled.
Who runs this
- Built and operated by Sensara LLC, Atlanta, Georgia — about us and how the audit works.
- See what the report looks like before you run anything — score per engine, the competitors AI names instead of you, and a fix plan.
- We run the same audit on ourselves every week and publish the result: in the latest run AI named AskedAbout in 1 of 144 answers. We report our own numbers the way we report yours.