Founder Decision Ledger #4: I Checked 45 International Outlets Against the Index. One Non-US Newsroom Clears the Evidence Floor.

Here is the pitch a founder with a London or Berlin launch on the roadmap gets handed: buy coverage in that market's press, because that is how the AI engines learn you exist there. It sounds right. The engines read the web, the market's newspapers are on the web, so the market's newspapers must be where the engines go when a buyer in that market asks a question.

I coined Machine Relations, I run the company that sells against these numbers, and I built the instrument that measures them. So the instrument gets to answer this one before I do. Today it did, and the answer is not the pitch.

This is entry four in the ledger. Entry one was whether my own site can win the worth question. Entries two and three were a platform and an engine. This one is a budget line.

Concede the part that is true

International coverage works on the people it was always meant to work on.

The Reuters Institute's 2026 Digital News Report, across 48 markets, finds trust in news at the lowest level it has recorded since 2015, and in the same breath finds that trust in the most widely used individual news brands is holding up better than trust in news overall. A buyer in Sydney who reads the Herald still weights the Herald. Cision's 2026 State of the Media, surveying 1,899 journalists across 19 markets, finds 66% of them rely on PR-provided material for story ideas, so the placement is winnable. None of that is in dispute.

What is in dispute is the second half of the pitch: that the placement is also how an AI answer reaches that market. That is a measurable claim, and I measured it.

What the Index says about nations

Machine Relations Research published the country-of-registration cut of the Index this morning: Country-Code Domains Take 2.7% of AI Citations. Every figure in this section is from that page and from the release it reads, mri_score_v2.0+2026-09-19+0cad03121f60, window 2026-05-10 to 2026-09-19, 15,883 answer runs, 22,213 cited domains, 111,510 domain-run citation pairs across six engines.

Where the cited domain is registered	Domains	Share of citation volume
Generic (.com, .org, .net and kin)	17,876	85.17%
Two-letter codes sold as brandable strings (.ai, .io, .co)	2,611	10.45%
National country codes, 91 territories	1,462	2.73%
US-restricted (.gov, .edu, .mil)	264	1.65%

The Index publishes a rate for a domain only after 10 observations across 7 distinct run dates. Three of the 1,462 national domains clear that floor, and all three are American businesses using .us as a short name. The count of non-US national domains above the floor is zero, out of 1,410.

The whole .uk zone, 337 cited domains, is 0.600% of citation volume. The whole .de zone, 96 domains, is 0.133%: 148 citation pairs out of 111,510 in 126 days, from the registry DENIC reports as one of the largest in the world. Japan contributed 22 domains and 34 pairs. Meanwhile .ai, delegated to the Government of Anguilla, and .io, delegated in the British Indian Ocean Territory, take 8.56% between them, which is 3.13 times every nation on Earth combined. Both are country codes under RFC 1591 and ICANN's definition. Neither says anything about where a company is.

That is the source layer. Now the outlets.

The 45 outlets, checked one by one

An international list is not an abstraction. It is a specific set of newsrooms an agency proposes, so I checked a specific set. I put together 38 non-US outlets a founder is routinely pitched for the UK, Europe, Asia-Pacific, Canada and India, plus 7 US outlets for scale, and read each one's public profile in the Index on 2026-09-19 at 13:52 UTC. The route is the same for every domain: machinerelations.ai/index/domains/<domain>. A profile lists every run in which the domain was cited, the days it was cited, the engines that cited it and its confidence grade. A 404 at that route means the domain is not among the 22,213 cited at least once in the window.

Here is every non-US outlet that has a profile, ordered by cited runs out of 15,883.

Outlet	Cited runs	Days cited (of 126)	Engines citing it (of 6)	Grade
indiatimes.com	37	24	5	C
theguardian.com	29	10	3	Collecting
thenextweb.com	20	14	5	Collecting
reuters.com	13	13	3	Collecting
independent.co.uk	12	8	2	Collecting
smh.com.au	10	6	1	Collecting
sifted.eu	7	7	4	Collecting
theglobeandmail.com	6	5	3	Collecting
bbc.co.uk	5	3	1	Collecting
tech.eu	5	5	2	Collecting
bbc.com	4	3	1	Collecting
ft.com	4	4	2	Collecting
lemonde.fr	4	4	1	Collecting
afr.com	4	4	1	Collecting
economist.com	3	3	2	Collecting
cbc.ca	3	2	2	Collecting
handelsblatt.com	2	2	1	Collecting
elpais.com	2	2	1	Collecting
thetimes.com	1	1	1	Collecting
eu-startups.com	1	1	1	Collecting
techinasia.com	1	1	1	Collecting

Twenty-one outlets, 173 cited runs between them, one grade. The other 17 on my list returned no profile at all on that date: aljazeera.com, corriere.it, dw.com, euronews.com, faz.net, france24.com, lesechos.fr, nikkei.com, scmp.com, spiegel.de, straitstimes.com, telegraph.co.uk, theaustralian.com.au, thehindu.com, thetimes.co.uk, wired.co.uk and zeit.de. Der Spiegel, Nikkei, the South China Morning Post and the Telegraph are not small newsrooms. Across 15,883 English-language commercial answers, the six engines did not cite any of them once.

Read the grade column. A grade is domain-level: the Index assigns one only once a domain has accumulated enough evidence across its segments, 503 of the 22,213 cited domains carry one in this release, and no graded domain has fewer than 30 cited runs. One non-US newsroom on this list has a grade: the Times of India's parent domain, at C, cited by five of the six engines. The Guardian is cited more often than any other non-US outlet bar one, on three engines, and still sits at Collecting, because its 29 runs are spread thinly across six categories.

For scale, the seven US outlets:

Outlet	Cited runs	Days cited	Engines	Grade
forbes.com	657	105	6	A
techcrunch.com	164	44	5	B
nytimes.com	141	31	3	B
cnbc.com	129	31	5	B
businessinsider.com	99	40	5	C
wsj.com	74	23	4	C
bloomberg.com	3	3	2	Collecting

Forbes alone is cited in 657 runs. The 21 non-US outlets with a profile are cited in 173 between them. One American business magazine outweighs the Guardian, the BBC, the FT, Reuters, Le Monde, the Globe and Mail, the Herald and the Times of India combined, by 3.8 to 1, in the answer layer. And Bloomberg, at 3 runs, is the reminder that this is not about prestige either. It is about which pages the engines actually retrieve when a buyer asks a commercial question, and the answer is not the newsroom's.

The news shape is the least international, not the most

If any part of the answer layer reached into national press, it should be the news-shaped question. The Index measures seven question shapes, and the national share of citations is lowest on exactly that one.

Question shape	Citation pairs	National-domain share
Problem-first research	8,856	4.55%
How buyers choose	11,242	3.93%
Ranked top lists	12,774	3.59%
Best tools	12,713	3.27%
Head-to-head comparison	9,915	2.97%
Is it worth it	10,794	2.81%
News-driven	45,216	1.61%

A problem-shaped buying question is 2.8 times more likely to reach a national domain than a news question about the same industry. This lines up with a finding Christian published this morning from the same release: the publications AI cites for an industry's news are not the ones it cites for its buyers. The international pitch assumes news coverage is the globally distributed surface and product content is the local one. It is backwards on both counts.

Category decides more than country

The same measurement splits by what you sell, and the split is the most useful thing in it.

Category	National-domain share	.ai and .io share
Emergent prosumer	7.15%	4.30%
Consumer health	6.91%	1.68%
Consumer products	6.27%	1.23%
Cybersecurity	1.93%	10.32%
Enterprise software	1.80%	8.75%
Fintech	1.64%	10.63%
AI visibility and GEO	2.44%	26.17%

Where the answer depends on a regulator, a health service or a physical product sold in a place, the engines reach into national infrastructure: nhs.uk, europa.eu, cdhf.ca, vic.gov.au. The single most-cited national domain in the release is europa.eu, cited by all six engines, and it is a government, not a publisher. Where the answer is about B2B software, national domains barely appear and the space is taken by .ai and .io vendor sites. In my own category, AI visibility and GEO, brandable two-letter domains take 26.17% of citation volume against a national share of 2.44%.

So a consumer-health founder and a fintech founder reading the same international pitch are being sold two different things. For one of them the national layer is already in the answer, and it is a regulator, not a newspaper. For the other the national layer is not in the answer at all.

The decision, and the four parts of it

One. International coverage comes out of the AI visibility budget and stays in the market-entry budget. It is bought for the humans in that market, who by the Reuters Institute's own numbers still weight the brands they use, and it is judged on what market entry is judged on: pipeline in that market. It does not get to carry a second justification it cannot earn. Zero non-US national domains above the evidence floor and one graded newsroom out of 38 is not a citation line. It was two line items pretending to be one.

Two. For citation outside the US, the money goes where the engines already reach for my category. That is the generic layer: the review databases, the community platforms, the aggregators and the .com and .ai sites the category's segment pages already show at the top. Open your category's segment page, for example fintech comparison answers, and read the top ten. Count the national domains. Then decide whether a placement in a national paper is going to change that list.

Three. No country-code domain for the citation surface. Google's own guidance treats a ccTLD as the strongest signal of country targeting. That is exactly the property that makes it invisible to an engine answering a question with no country in it: the entire .de zone contributed 148 of 111,510 citation pairs. A brand that needs both keeps localized versions of its pages on the domain the global layer already cites.

Four. I do not fund a claim the instrument cannot see. This is the part of the pitch I will not pretend is settled either way. The Index observes a fixed basket of English-language commercial questions with no country in them. It cannot tell me what a German buyer asking in German sees, and this page makes no claim about that. Until a locale-varying basket exists, "AI reaches that market" is unmeasured, which is not the same as zero. Unmeasured does not get the visibility line. It gets a measurement first.

The ledger entry

Every decision I make from this data gets recorded the same way, including the condition that kills it.

Field	Entry
Decision	International press coverage is a market-entry cost, not an AI visibility line; AI citation spend outside the US goes to the generic source layer the category's segment pages already show
Date	2026-09-19
Evidence	MRI release mri_score_v2.0+2026-09-19+0cad03121f60, window 2026-05-10 to 2026-09-19, 15,883 runs, 22,213 domains; the country-code study; the 45-outlet profile census read 2026-09-19 13:52 UTC
Scope	English-language commercial questions with no country in the prompt, six engines, 126 days; not a claim about localized queries in any market
Reversal condition	Any release where 10 or more non-US national domains clear the evidence floor (today: 0 of 1,410), or where any outlet on the 38 above other than indiatimes.com carries grade B or better, or a locale-varying measurement shows a national-domain share above 10% for a market I sell into
Review	Next quarterly release read, with the same 45 profiles re-read at the same routes

If that reversal condition fires, I was wrong and I will say so here.

What this measurement is and is not

The Index observes six engines answering the market's buyer questions: ChatGPT, Claude, Gemini, Google AI Mode, Google AI Overviews and Perplexity. This release covers 15,883 answer runs and 22,213 cited domains over 126 days, and the machine-readable release carries every number above. Classification by country is by the last label of the domain name, which is what a registry and a resolver both use; the split between national codes and brandable two-letter codes is an editorial judgment the study reports both ways, and the strict reading, where every two-letter domain is national, still puts .ai, .io and .co at 75.5% of that group.

Three limits matter. The release carries no locale, language or region field, so every figure here is about an English question with no country in it and nothing else. A domain with no profile was not observed in this basket; that is a statement about 15,883 runs, not about the outlet. And citation is correlation: nothing here says a placement caused or failed to cause anything, and nobody should publish that it does.

The binary

You are buying one of two things when you buy international coverage. Either you are buying the attention of readers in a market, which is real and which you should judge as market entry. Or you are buying a route into the answer layer for that market, which nobody has measured, and which the only instrument that measures the adjacent question says runs through .com and Anguilla, not through the national press.

Fund the first one honestly. Measure the second one before you fund it at all.

FAQ

Does this mean international PR is worthless? No. It means it is worth what it is worth to the humans who read it, and it should be budgeted and judged as market entry. The evidence here does not support charging it to AI visibility as well.

Would a placement in the Guardian or the FT help my AI citations? Across 15,883 English commercial answers the Guardian was cited in 29 runs and the FT in 4, and neither has accumulated a publishable rate in any category. Forbes was cited in 657. A placement is worth pursuing for its readers; on this evidence it is not a citation strategy.

My buyers are in Germany. Does the Index say what they see? No, and it says so. The Index runs English questions with no country in them. The .de zone's 148 citation pairs describe that basket, not a German-language query in Germany. That is the measurement to run before you spend.

Should I move my site to a country-code domain for a new market? Not for citation. National domains are cited once, by one engine, on one day: 86.5% of them were cited by exactly one engine and 61.6% in exactly one run. Keep localized pages on the domain the engines already cite.

How do I check an outlet myself? Open machinerelations.ai/index/domains/ followed by the outlet's root domain. The profile shows cited runs, days cited, the engines that cited it and its grade, with the release id at the top. A 404 means the domain was not cited in the current release.

Can I reproduce the census? Yes. The 45 domains are listed above, the routes are public, and the release id fixes the version. Read them on the same day and the numbers will match; read them on a later release and any difference is the next entry in this ledger.