There Is No Shortlist of Publishers: 58 of 67 Cited Domains Appear in One Category Only

The advice everyone gives about AI visibility is to get published on the sites AI already trusts. So we counted them. Across twelve categories in the United States, sixty-seven domains hold a durable citation and fifty-eight of them appear in exactly one category. There is no list.

One domain spans categories in every market we measured, and it is YouTube.

Disclosure: EchoWi sells AI visibility measurement, so a study concluding that placement is harder than it sounds is a study with an interest. This one is derived rather than newly measured: every number below is recomputed from rows already published in our measurement register, with the query, market, surface, date and run count on each. Anyone can do the arithmetic against those rows and get a different answer if we made a mistake.

The short version

1. Fifty-eight of sixty-seven domains appear in one category only. Those are the domains that held a citation in every run of a category’s buying question in the United States.
2. Nine span more than one, and the top of that list is not a publisher. It is YouTube, in eight categories of twelve.
3. Reddit and Zapier are next, at four each. Nothing else reaches four.
4. The pattern survives leaving the United States. In Spain twenty-eight of thirty domains are category specific, in Germany seventeen of twenty-one, and YouTube leads both.
5. Two of the nine spanning domains are vendors, not publishers, and both sell into the adjacent categories they appear in.
6. This is a floor, not a ceiling, and the section on limits explains why the true amount of spanning is higher than these counts and still nowhere near a shortlist.

What was counted

Our published work measures one category at a time: ask the buying question, repeat it, and record which domains came back in every run rather than once. That produces a stable core per category.

This study puts those cores side by side and asks a question none of them could: how many of these domains hold a slot in more than one category?

Twelve category questions in the United States, five in Spain, three in Germany, all on Google’s AI Overview with country and language fixed, measured between 9 and 11 August 2026. One of the twelve is a non-comparative control on ecommerce rather than a twelfth sector, and it is left in the count because removing a row after seeing which way it pushed the answer is how a study talks itself into a number. Where a category was measured more than once, the first reading is used, because counting a category twice would manufacture exactly the concentration the study is looking for.

The United States

The nine:

Read the top of that table again. The domain that holds a durable citation in two thirds of the categories we measured is a video platform, and the second and third are a forum and an integrations directory. The publications a content plan would name, PCMag and Forbes and CNBC, are in two or three each.

Spain and Germany

The shape is the same, and the sample is smaller so the counts are smaller.

In Spain the two that span are YouTube, in 5 categories of 7, and Reddit in 3. In Germany 4 span: YouTube and fuer-gruender.de, a portal for people starting a business, both in 5 of 7, then handelsblatt.com in 3 and omr.com in 2.

So the one domain that spans categories in all three markets is YouTube, and everything else is either category specific, market specific, or both. That matches what we have found every other time we have crossed a border: the ordering of categories travels and the names do not.

The two vendors on the list

Salesforce holds a durable slot in three categories: CRM, which it sells, email marketing, which it sells, and help desk, which it sells. QuickBooks holds one in accounting and in payroll, and Intuit sells both.

That is worth separating from the rest of the table, because it is the only route in it that a company can actually take. Neither of them got there by being a big brand in a neighbouring category. They got there by being a product in that category with pages about it, and the answer to “best help desk software” cited Salesforce the way it cited any other vendor with a real product there.

It also sets the ceiling honestly. Two vendors span, out of sixty-seven domains, and both did it by selling more than one thing rather than by publishing more than one thing.

What this means for a content plan

The “get published where AI already looks” advice does not survive contact with the list. It is good advice in the sense that placement matters, and thin in the sense that there is almost no shared destination: eighty-seven per cent of the domains holding a slot hold it in one category. Whatever the destination is for a category you have not measured, it is not a list somebody can hand you.

The one universal is video. YouTube is in eight of twelve US categories, 5 of 7 Spanish and 5 of 7 German. We would not turn that into “make videos” until we knew whose videos were being cited, and that has now been measured: across four of these categories, ten cited videos belong to review publishers and independent creators and none to a vendor in the category. So this is a placement problem like the rest of the list, not a production one.

The forum is the second universal and it is not buyable. Reddit is in four US categories and three Spanish ones. There is a whole industry selling placement in it and no honest way to describe that as publishing.

And the practical consequence is unglamorous. If you want to know which domains hold the answer in your category, in your market, the only way to find out is to ask that question and look at the sources. That is what this study is made of, twenty times over, and it is the only step that was not guesswork.

The same question on the other surface, with the bar taken off

Everything above is AI Overview, and it counts only the domains a category cited in every run. Both are choices that could be producing the answer. A strict bar keeps a domain out of a category, and a domain kept out of two cannot span them.

So the count was run again with both choices reversed, over twelve cells another study had already paid for: AI Mode instead of AI Overview, and every domain a category cited at all instead of only the durable ones. Six categories in the United States and the same six in Spain, on 13 and 14 August 2026.

That is the permissive direction on both axes. Counting every cited domain rather than the durable ones multiplies the population, and more domains counted is more chances for one to appear twice. If a shortlist were hiding behind the strict bar, this is where it would show.

It does not show. YouTube is in all 6 of the United States categories and 5 of the 6 in Spain, which is the same domain at the top of the same list as on the other surface. Behind it the United States has Reddit in 5, Google itself in 5, and PCMag, the US Chamber of Commerce and Forbes in 4 each. Spain has one domain in 4 and six more in 3.

Almost nothing crosses the border

The count above says that whatever does not span categories is category specific, market specific, or both, and it never puts a figure on the second, because its markets were measured with different category sets. These twelve cells are the same six categories in both, so the comparison is finally like for like.

Of 451 distinct domains cited across the two markets, 13 appear in both. Under three in a hundred.

The thirteen, by name: allaboutcookies.org, bitwarden.com, bluehost.com, cnet.com, fastlancer.org, google.com, hostinger.com, minea.com, seedprod.com, stripe.com, youtube.com, zdnet.com and zoho.com.

The list is published rather than summarised because we did not open these thirteen home pages this pass, and a domain name is a hypothesis about what a site sells rather than the answer. Read the list and check it. What we will say without classifying anything is the shape it implies for a content plan: an answer layer that is national nearly all the way down, with a thin spine of domains that appear on both sides of it.

What those thirteen turned out to be

The list above was published without a classification, because a domain name is a hypothesis about what a site sells and its home page is the answer. Each of the thirteen was then opened, on 14 August 2026, and recorded by what its own page says.

7 of the 13 border-crossing domains are the vendors themselves. Bitwarden, Bluehost, Hostinger, Minea, SeedProd, Stripe and Zoho all sell the product their category’s question is about, or a product next to it. Two are comparison publishers, allaboutcookies.org and fastlancer.org. One is a platform, YouTube. One is Google citing itself, which this register counts separately and does not treat as a publisher.

Two answered a bot challenge rather than a page, cnet.com and zdnet.com, and are recorded as unclassified rather than guessed. Both are plausibly technology press. If they are, the split is 7 vendors to 4 publishers rather than 7 to 2, and the finding below does not change either way.

So the layer that crosses a border is mostly not a publisher layer at all. It is vendors on their own domains, plus YouTube.

That is worth stating plainly because it inverts the obvious content plan. In one market the work is to be cited by the publishers of that market, and there are hundreds of them and almost none of them reach the next market. Across markets, the thing that travels is your own domain.

One detail rewards a second look. fastlancer.org is a German-language comparison site, and it appears in both the English-language United States answers and the Spanish-language Spanish ones, while being the national publisher of neither. We have no explanation for that and are not going to invent one; it is in the register by name so somebody can check it.

What this run cannot say. Two things changed at once, the surface and the bar, so this corroborates the shape and cannot attribute any difference to either. Depth also differs, 41 answered runs in the United States against 36 in Spain, and Spain still cites more distinct domains, so the larger Spanish set is not an artefact of reading it deeper. Six categories per market is half the twelve counted above. And google.com in five United States categories is the engine citing itself, which we measure separately and do not treat as a publisher.

What this does not show

The spanning count is a floor. A domain serving two categories only shows up here as spanning if we happened to measure both. Twelve categories out of a universe far larger than twelve makes that unlikely, so the true amount of overlap is higher than nine. What the floor does establish is the shape: even inside a sample chosen to be broad, the overlap is concentrated in a video platform and a forum.

One surface. Google’s AI Overview. Our own work says source sets do not transfer between surfaces, so a domain absent here may hold a slot on another.

Durable means within a session. A domain counts if it came back in every run of that category’s question on that day. That measures stability inside a reading, not stability over weeks, and we have published a case where a three of three became two of three in a single day. It is also why this study counts domains rather than ranking categories by their stable share: pooling every measurement in the register shows that the share largely reports how many sources the answer cited, which is a property of the question rather than of the category.

Twelve categories, five, three. The Spanish and German samples are small enough that their counts should be read as direction rather than as measurements.

Nothing here is causal. No site changed anything and we observed the result. These are the sources a surface assembled answers from on those days.

Common Questions About Getting Cited

Is there a list of sites I should get published on to appear in AI answers?

Not in our data. Across twelve categories in the United States, fifty-eight of the sixty-seven domains holding a durable citation appeared in exactly one of them. The domains that span are a video platform, a forum, an integrations directory and two vendors selling into adjacent categories. If someone is selling you placement on “the sites AI trusts”, ask which categories they measured that in, because the answer to your category is probably not on their list.

Why is YouTube cited so much?

We can tell you that it is, in eight of twelve US categories, four of five in Spain and two of three in Germany, and we cannot tell you why from this data. What we deliberately have not claimed is that you should therefore make videos, because we have not measured whether the cited videos belong to brands or to independent reviewers. Those two worlds imply completely different work and the difference is one measurement away.

Does being big in one category help me in the next one?

Only two domains in this data span categories as vendors, Salesforce and QuickBooks, and both sell real products in every category they appear in. Neither shows up in a category it does not compete in. So the honest reading is that product presence travels and brand size on its own does not, at least not far enough to hold a durable citation.

How many categories do I need to measure to know my own?

One, and it has to be yours. That is the least satisfying answer in this article and it is the one the data supports. Categories do share a few domains, mostly YouTube and Reddit, so the claim is not that the cores are disjoint. It is that 58 of the 67 domains holding a slot appear in a single category, so a neighbouring category’s source list gets you the two you could have guessed and misses the rest.

Do these lists change over time?

We do not know yet, and anyone who tells you otherwise has not published the check. Every reading here is from a three-day window, so this measures which domains are stable within a session rather than across months. We have one case published where a citation went from every run to two of three in a single day, so the honest expectation is that the core is stable and its edges are not.

What should I do first with this?

Ask your category’s buying question, repeat it three times with the country and language fixed, and write down which domains came back every time. That list is the only one that describes your market, it takes minutes, and it will tell you whether you are looking at a category where vendors hold slots or one where nobody does. Everything after that is a placement problem with a known target instead of a guess.