One Domain Is the Best Agency in Five Different Industries. The Other 123 Are in One.

We asked seven industries the same question, “what is the best marketing agency” for that sector, on three AI surfaces with the cache bypassed. 148 domain citations came back, 133 of them distinct, and only ten appeared in more than one industry. Ninety-two per cent of what an assistant cites for one sector, it cites for no other.

That is the finding, and it decides something practical: there is no such thing as being cited as “a good agency”. You are cited as a good healthcare agency or a good hotel agency, and the two are almost entirely different lists.

Disclosure: EchoWi sells AI visibility measurement, so several domains below are competitors and one of the seven questions is about the category we sell into. That row has a conflict the others do not, and the prompt, market and run count behind it are published like all the rest, so it can be checked separately. Every row is in our measurement register with its question, its market and its run count.

The short version

1. 133 distinct domains, 92% exclusive to one sector. Ten domains crossed sectors at all.
2. One domain crossed five of the seven. marketingltb.com was cited as the answer for law firms, hotels, white label SEO and higher education, at every run in four of them.
3. Law firms is the stable outlier, at 65% against 31% for the other six sectors pooled, and the difference survives a test rather than resting on the gap looking large. It does not survive a comparison against the single next-highest sector, which is the honest qualification and is worked below.
4. Reddit is cited in three sectors, and is not an agency.
5. The stable half is smaller than it looks. Outside law firms, between 26% and 37% of each sector’s cited set survived every run.

How this was measured

Seven questions, one per sector, phrased the way a buyer types them. Three surfaces: Google AI Overview, Google AI Mode and Gemini. United States, English, both set explicitly. Measured on 9 August 2026.

Every run bypassed the cache and was billed individually. This matters more than it sounds. Repeating a prompt against a cached endpoint returns the same text and produces a denominator that was never measured, so a rate built that way is one observation wearing a percentage. These are 51 real calls.

The run counts are what completed, not what we asked for. We requested three runs per surface on all seven. Four sectors stopped at two and said so. Those rows say two, because writing three would invent a third of a denominator, and this register exists precisely to stop that.

What came back

Does that outlier survive an interval?

Our own rule for a small sample asks for counts, a distribution and an interval. This piece published the first two and not the third, so here it is, and it changes which comparison the claim is allowed to rest on.

At 95%, law firms sits at 65%, in a band from 43 to 82. The next-highest sector is B2B SaaS at 37%, in a band reaching 59. Those bands overlap, so law firms is not separable from the single next-best sector, and anyone reading the chart as a ranking of seven sectors is reading more than the data holds.

Against the other six pooled, it is a different matter. The rest together are 31%, in a band from 24 to 40, which does not overlap 43 to 82, and a two-proportion test gives z = 2.93.

One confound is worth closing rather than mentioning. A unanimity bar over two runs is easier to clear than over three, this register has measured that elsewhere, and law firms is a two-run cell. Comparing it only against the other two-run cells, so the bar is identical on both sides, gives z = 2.56, which still separates at 95%.

So the finding stands and the wording had to move: law firms differs from the field, not from its nearest neighbour.

The ten domains that crossed sectors

Out of 133, these are every domain cited for more than one industry:

Three of those ten are not agencies at all. Reddit, YouTube and LinkedIn are places where people talk about agencies, and they are cited alongside the agencies themselves.

The other 123 domains appear in exactly one sector. That is the number to plan against.

We ran it again in Spain, and the borders are not where we expected

The finding above says sectors do not share sources. The obvious next question is whether markets do, and our three-market study had already found stable citation sets that do not overlap between countries. So we asked four of the same questions again in Spain, in Spanish, with country and language set explicitly.

The overlap between the Spanish answer and the American one, for the same sector, runs from 44% to nothing.

Hotels behave like one global market and law firms like fifty separate ones. The shared hotel domains are the sector’s international trade press and international agencies: hoteltechreport.com, d-edge.com, propeller.co.uk, amraandelma.com, revolutionmedia.agency, americasgreatresorts.net, and marketingltb.com again. Every domain the Spanish law answer cited was Spanish, and not one of them appeared in the American answer.

That is not a surprise once stated, because a hotel competes for a traveller who could come from anywhere and a law firm competes inside one jurisdiction. It is worth measuring anyway, because it changes what a piece of English content is worth: in hotels, close to half of what you win carries; in legal, none of it does.

The healthcare overlap is two domains, and one of them is YouTube.

A caution on the denominators. Two of the four Spanish questions completed three runs per surface and the hotel one completed two. On the law-firm question, AI Mode and Gemini answered three times and AI Overview twice. These are enough to separate a stable citation from a coincidence and not enough to call any of them a rate.

We tested that in France and Germany, and it holds

If a sector’s citation set travels when its customers do, hotels should share sources across borders and law firms should not. Hotels and law firms were the two extremes of the Spanish result, so they are the two that could falsify it. We asked both again in France and in Germany.

Hotels have an international core and law firms have none. Five domains are cited for hotels in the United States, Spain and France alike: hoteltechreport.com, d-edge.com, revolutionmedia.agency, amraandelma.com and americasgreatresorts.net. Two of those, the trade publication and the booking-technology vendor, are cited in all four markets.

For law firms the number is zero in every direction. Not one domain is cited for law firms in the United States, Spain and France together. Across all four markets the only cross-border pairs at all are mec-digital.com, cited in Spain and France, and natlawreview.com, cited in the United States and once in Germany. Four national markets, four separate answers.

Germany is the finding we did not expect. It behaves nationally even in the global sector: German hotels share two domains with the American answer and three with the French, where the United States, Spain and France share five to seven with each other. Whatever makes hotel citations travel, it travels less into German.

That is a caution for us and for anyone planning market entry on the back of an English page: the sector predicts whether your citations cross a border, and Germany is a border the rest of this data did not prepare us for.

What it means if you sell services

A generic agency page is not a citation strategy. The evidence for that is not an opinion, it is that 92% of the cited domains in this measurement are sector-exclusive. Whatever an assistant is doing when it assembles these answers, it is not consulting one canonical list of good agencies and filtering it. It is building a different list per sector.

Being cited in one sector tells you almost nothing about the next. If you win healthcare, you have won healthcare. The hotel list shares one domain with it, and that domain is YouTube.

The stable half is smaller than it looks. Outside law firms, between 26% and 37% of each sector’s cited set survived every run. The rest is a different plausible subset each time, which is the same instability we have measured in category answers generally. Anyone reporting a sector visibility score from one check is reporting a draw.

The exception worth studying

marketingltb.com was cited as the best agency for law firms, hotels, white label SEO and higher education, on every run of each, plus ecommerce on one run in three. One domain, five industries, in a measurement where 92% of domains manage one.

We are not going to tell you why from outside, because we did not measure why. What we can say is that it is the only domain in this set that behaves like a general answer to a sector-specific question, and that whatever produces that behaviour is worth more than another comparison table. It is on our list to look at properly.

What we could not verify

• Why any domain is cited. This measures what came back, not the mechanism.
• Whether these lists hold next week. Four sectors ran twice and three ran three times. That is enough to separate a stable citation from a coincidence and not enough to call it a trend.
• Other markets. United States and Spain. France and Germany are unmeasured, and having seen two markets produce overlaps ranging from 44% to zero, we would not assume the result in a third.
• ChatGPT. Left out on purpose: at three runs across seven questions it stops early often enough that the denominators would have been ragged.

Common Questions About Vertical AI Citation

Do AI assistants recommend the same agencies across industries?

Almost never. In this measurement, 133 distinct domains were cited across seven sectors and only ten appeared in more than one. Ninety-two per cent were cited for a single sector.

How many sources does an AI answer about agencies use?

Between 17 and 25 distinct domains per sector across two or three runs on three surfaces. The larger number is not a better answer: white label SEO returned the most domains, 25, and only 8 of them came back on every run.

Is a single AI visibility check enough to report a sector score?

No. Outside law firms, roughly a third of each sector’s cited domains appeared in every run, so two thirds of any single check is a set that will look different next time. Rates need repeated, uncached runs.

Which non-agency sites get cited as agency recommendations?

Reddit in three sectors, YouTube and LinkedIn in two each. They are cited alongside the agencies rather than instead of them, and they are among the few domains that cross sectors at all.