YouTube Is Cited in Two Thirds of Categories. We Checked Whose Channels.

Counting across categories, YouTube holds a durable citation in more of them than any other domain, in every market we measured. We said at the time that this was not advice to make videos, because nobody had checked whose videos they were. We have now checked, twice. In the first reading, ten videos and eight channels with not one belonging to a company in the category being compared. Widening it to 25 videos in four markets, 20 channels and exactly 1 that does.

They belong to review publishers and to individual creators, and the difference matters more than the platform does.

Disclosure: EchoWi sells AI visibility measurement, so a study concluding that video is a placement problem rather than a production one is a study with an interest. The four questions, the ten video IDs and every channel name are in our measurement register with the market and the date, and so are the 12 later cells with all 25 of their video IDs, the channel behind each one, the same 12 cells read again on the other surface, and 12 more cells in 3 further categories. Each channel was resolved by loading the video and reading its owner, not by inferring from the title, and anyone can repeat that in a browser.

The short version

1. Zero of the first ten cited videos belong to a vendor in the category, and 1 of the 20 channels in the wider reading does. Not HubSpot in CRM, not QuickBooks in accounting, not Mailchimp in email marketing, not 1Password in password managers.
2. Seven belong to individual creators and three to review publishers.
3. Eight distinct channels for ten videos. Two channels were cited twice, so there is a little concentration and not much.
4. Every title is comparative. “Top 5”, “Tested 15 CRMs”, “TIER LIST”, “Xero vs QuickBooks”. The same shape that wins in text wins in video.
5. The channels are small and reachable, which is the one piece of good news here and the reason this is worth acting on.
6. This is one draw per category, not a rate, and the method section says exactly what that does and does not support.
7. No video crosses a border. Asked again in 3 categories across 4 markets, 12 cells cited 25 distinct videos and 0 of them appear in two markets. The slot is real everywhere and it is a different video everywhere.
8. And it does not cross a surface either. The same cells on AI Mode share 8 videos with their own AI Overview answers, an overlap of 0.18, and 7 of the 12 share none.
9. A vendor channel is a category thing, not a rare thing. Widening to 3 more categories doubled the channels and turned 1 vendor into 4 of 32, and they cluster: 2 in email marketing, 1 in accounting, 0 in password managers.

What was measured, and with which tool

Four categories where our earlier work found YouTube holding a durable slot: CRM, accounting software, email marketing and password managers. Each was asked its buying question once on Google’s AI Overview, United States, English, on 11 August 2026.

The tool matters and it is not the one we normally use. Our own rule is that any figure published as a rate has to come from repeated runs with the cache bypassed, because a single run is a draw. This study publishes no rate. The quantity is which videos were cited and who owns those channels, and identity is exactly what a single cached read is for.

So there is no percentage anywhere below. Ten videos were cited across four answers and zero of them are vendor channels. If the same questions are asked tomorrow the ten may be a different ten; what this establishes is what kind of channel occupies that slot, and one draw across four categories is enough to say that a vendor channel did not occupy any of them.

Every channel was resolved by loading the video and reading the owner field. A title like “Top 5 CRM Software For Small Businesses” tells you nothing about who made it, and who made it is the entire question.

The ten videos

Read the channel column. Two of the eight are publications with a review business, Merchant Maverick and VPNpro, and the other six are people. Stewart Gauld is cited twice in accounting, once for a top five and once for a head to head.

Why this kills “so make videos”

The obvious reading of a domain holding slots in two thirds of categories is that the platform is the opportunity. If a brand’s own channel were among these ten, that reading would be right and the work would be production: film a demo, publish it, get cited.

None of them is a brand’s channel, so the work is not production. It is the same problem the text layer has, which we have measured repeatedly: the pages that win a category question compare products, and a vendor’s own pages describe one. A vendor’s video does the same thing its pricing page does. It describes one product, and the question asked for a comparison.

That is why the titles are worth reading as data rather than as decoration. Ten videos and every one of them is a ranking, a tier list, a top five or a head to head. Not one is a product tour.

The format is the same one that wins in text

Put the ten titles next to the written sources cited in the same four answers and the pattern is one pattern, not two.

The written sources are PCMag’s tested roundups, Zapier’s “8 best free email marketing services”, Venture Harbour’s “10 best email marketing software (+ 3 I’d avoid)”, the US Chamber’s “10 free accounting tools”, Gartner’s market page and Reddit threads asking which one to use. The video sources are “Tested 15 CRMs”, “Top 5 Accounting Software”, “Best password manager TIER LIST”, “Xero vs QuickBooks”.

Every one of those, in both media, does the same job: it holds several products against each other and reaches a ranking. Nothing in either list describes a single product except the vendor pages that appear alongside, and those show up when the answer needs a specific fact such as a price rather than when it needs the recommendation.

That is worth stating plainly because it collapses two pieces of advice into one. “Get published on the sites AI trusts” and “make videos for AI search” are usually sold as different projects with different budgets. On this evidence they are the same project: be inside comparative content, wherever that content lives. The platform is a detail of where the comparison happens, and the comparison is the thing being cited.

It also explains the one exception we have already published. When the question names a brand and asks what it costs, the brand’s own page is cited, because at that point the answer needs a fact about one product and the vendor is the authority on it. Comparison questions and fact questions are answered from different kinds of source, and the video layer does not change that division. It follows it.

The part that is actually good news

Six of the eight channels are individuals or small niche channels. Consumer Desire, CRM Central, Daniel, Stewart Gauld, Jamie Trull, Craylor.

Compare that with the text side of the same answers, where the sources are PCMag, Wired, Forbes, Gartner and the New York Times. Getting into a Wirecutter roundup is a project. Getting a working relationship with a creator who makes accounting software comparisons and is already cited twice in one answer is an email.

We are not going to tell you that outreach to those creators produces citations, because we have not measured that and nobody has published a controlled test of it. What the data supports is narrower and still useful: the slot exists, it is held by people rather than institutions, and those people are reachable in a way the institutions are not.

One video, or one video per market

The section above is only an opportunity if one video can serve more than one market. If the slot is a video, entering it is one production. If it is a video per market, it is four, and that is a different decision with a different budget. Nobody had measured which, us included, because the counting that found YouTube in most categories counted the domain and never the video.

3 categories asked in 4 markets, each in its own language, on the same surface, one reading each: CRM, web hosting and ecommerce platforms. That is 12 cells, and every one of the 12 cited at least one YouTube video.

Those 12 answers cite 26 URLs, which resolve to 25 videos, and 0 of the 25 appear in two markets. 0 of the 25 appear in two categories either.

The unit is the whole reason the first line of that paragraph has two numbers in it, and the very first cell read is why. Six YouTube URLs in one citation list were five videos: one video was cited twice, once with a timestamp and once without, and one of the six was a Shorts link. Counting URLs would have published 26 videos with one of them counted twice. We have already paid once here for calling URLs pages, and this is the same mistake one level down.

The corroboration is free and it points the same way: in all 12 cells the cited videos came back with a transcript in that market’s own language. This is not one language-neutral video travelling between countries. It is a local one each time.

One instrument note belongs with that, because it would have fooled a shortcut. 2 of the 25 videos came back carrying an English title over a transcript that was not English. Anyone sorting these by title would have filed them as English-language videos appearing in a Spanish answer. The transcript is the evidence and the title is not.

One more thing the citation list does not carry. In 1 of the 12 cells the answer printed a video card in its body whose URL is absent from the list we counted from, so that cell cited a video we did not count. It can only push the video count up, and the video it adds appears nowhere else, so it cannot create a crossing. We say it because a count taken from a citation list is a floor and the answer above it is the thing a buyer sees.

The one cell where we already had a before

The 4 CRM videos in the table above were read on this same surface and market on 11 August, in question form. This arm read the keyword form of the same category on 17 August, and returned 5 videos, of which 3 are the same 3.

That is one observation and not a rate, and it is deliberately not put next to the overlap figures elsewhere in this register, which are durable-source overlaps measured under a unanimity bar across repeated runs. This is a single cached read of which videos were named. What it supports is narrow: across two question forms and six days, most of that cell’s videos were still there.

And the channel, which is the half that decides the cost

We wrote above that we had not resolved ownership. We have now, on all 25 videos, by loading each one and reading the channel that published it.

The 25 videos belong to 20 channels, and 0 of those channels appear in two markets. So the four productions the section above prices are four relationships as well. Per market the channels are 7, 6, 3 and 4, and 4 channels hold more than one of the 25.

What does cross is the other axis. 2 channels hold a slot in two categories of their own market: a German review publisher appears in both hosting and ecommerce, and a French one in both CRM and ecommerce. Inside one market, a channel that has earned the slot in one category can hold it in the next.

And one publisher does span two markets, in a way that makes the point sharper rather than weaker. The French hosting slot is held by Cybernews en Français and the Spanish hosting card by Cybernews en Español: the same publication, two channels, one per language. The brand crosses the border and the channel does not, which is exactly the shape of the cost.

We doubled the channels, and one vendor became four

“1 of 20 channels is a vendor” is a sentence about channels, so the thing to widen is channels. Three more categories in the same 4 markets, same shape, same method: email marketing, accounting software and password managers.

That is 12 more cells. 2 of them returned no AI Overview at all, both password managers, in Spain and France, which is the same category and the same two markets our render study had already measured as short, on a different wording and a different day. 9 of the 12 cited a video, giving 17 videos and 16 channels.

3 of those 16 channels belong to vendors selling in the category they are cited for, and the first arm’s one was an understatement. Across both sets it is 4 vendor channels out of 32.

They are not spread evenly, and that is the finding. Both email marketing vendors publish a channel that reviews the category they compete in, one in Spanish and one in German. The accounting vendor does the same in Spanish. The three categories from the first set produced one between them, and password managers produced none.

So the honest version is not “a vendor channel is rare”. It is that a vendor channel is something some categories have and others do not, and email marketing is the one where we found two.

The brand crosses the border, three times now

The first arm found one publication running a channel per language. Widening found two more: a review publisher and a creator, each with a separate channel for another market. 3 brands, 3 pairs of channels, and in every pair the videos differ.

That is the same shape from a bigger sample and it sharpens the cost sentence rather than softening it. The people who hold this slot do work per language, so the relationship you can build is per language too.

The vendor video the answer showed and the list left out

One more thing that is worth seeing and not worth counting. The German accounting answer printed a video card for an accounting vendor’s own channel, and that URL is absent from the citation list the counts here come from.

It is excluded, as every card-only video is in this study. It is reported because a vendor video sitting in the answer body is exactly the thing a vendor would want to know is possible, and because the count above would have been 4 of 17 rather than 3 of 16 if we had quietly folded it in.

And it does not survive a change of surface

Everything above is Google’s AI Overview. The limits below say so, and this register has measured the surface as the largest source of disagreement it knows: two surfaces of the same provider, asked the same question on the same day, share less of their stable sources than two readings a week apart do. So the price above is a price for one surface until somebody checks.

The same 12 cells, same strings, same markets, same day, on AI Mode.

11 of the 12 cited a video, the two surfaces name 25 and 27 videos, and 8 of them are the same. That is an overlap of 0.18, and 7 of the 12 cells share no video at all with their own AI Overview answer.

The clearest cell is French hosting. AI Overview cited a video for that question; AI Mode answered the same string on the same day with 31 sources and not one of them was a video.

So the advice gets a second dimension it did not have this morning. A video is a placement in one market and one surface. If the answer you care about is AI Mode, the twenty five videos above are largely not the ones holding it.

Both surfaces show a video they do not list, and neither direction is safe

Two AI Mode answers printed a numbered citation marker for a video their returned list left out. 1 of those 2 videos is an AI Overview video of the same cell and 1 is not, so the omission does not push the overlap up or down in any direction anyone can name.

It runs the other way too. The one video AI Overview printed as a card and left out of its own list is cited by AI Mode, in the Spanish hosting cell, which is the only shared video that cell has and it is invisible in the count above.

This is stated as two counts rather than as a correction because a caveat that names a direction is a claim, and this register has published one of those and had to withdraw it. What both counts support is narrower and enough: an overlap measured from citation lists is measured through two different holes, and the holes are not the same shape.

The vendor channel we said did not exist

The earlier study opened ten videos and found none belonging to a company in the category being compared. Reading a lot into that would have been wrong, and this arm shows why: of these 20 channels, 1 belongs to a vendor selling in the category its video is cited for. It is a Spanish CRM vendor whose own channel published a ranked list of the seven best CRMs for small business, and the answer cites it.

That is the format this study said wins, made by a vendor about its own category, and cited. It does not overturn “zero of ten”, because that count was correct for those ten. It bounds it: a vendor channel in this slot is rare in what we have measured and it is not impossible, and the one that made it did not publish a product tour.

3 more of the 20 belong to companies that sell something real and not the thing being compared: a marketing agency, a CRM consultancy and a freelance-recruitment platform. They are counted in their own class rather than folded into creators, because folding them in would have inflated the number this section exists to report. The remaining 16 are 7 review publishers and 9 independent creators.

The publisher that holds both slots in one answer

In 2 of the 12 cells the same publisher is cited twice in one answer, once as a page and once as a video. The German hosting answer cites a review site and that site’s own YouTube channel; the German ecommerce answer does the same with a different publisher.

That is worth separating from everything else here because it is the only route we have measured that buys both slots at once, and it is not available to a brand by imitation. Both are publications that already had the text citation and added video to the same brand.

What it costs, then

The good news above keeps its shape and gains a price. The slot is held by people rather than institutions, and it is held by a different video in every market we asked. One video is one market’s placement.

What this arm cannot tell you is whether it is also a different channel. We resolved video identity across all 12 cells and we did not resolve ownership for all 25, so a single channel spanning two markets is still possible and would change the cost again. What is closed is the cheaper hope: a single video does not span them.

What this does not show

One draw per category. Four questions, asked once each. This identifies who holds the slot, not how often any individual video holds it. A second reading would very likely return a different mix of videos and, on this evidence, the same kind of channel.

Four categories. CRM, accounting, email marketing and password managers, chosen because our earlier counting found YouTube durable in them. It is not a random sample of categories and does not claim to be.

Two surfaces, and every count belongs to an arm. The ten videos and every channel name are Google’s AI Overview, United States, English; the crossing arm is 4 markets on AI Overview and resolves both the videos and their channels; the surface arm repeats those same cells on AI Mode and resolves videos only; the widening arm adds 3 more categories on AI Overview and resolves channels again. We have published that source lists do not transfer between surfaces, so nothing here says anything about the other surfaces, and the ten names are the least transferable part of this.

Zero is a count, not a rate. Ten videos with no vendor channel among them does not mean a vendor channel is never cited. It means it did not happen in these four answers, which is enough to stop treating “YouTube gets cited” as a reason to publish product videos and not enough to say it never works.

Nothing here is causal. No channel changed anything and we watched what happened. These are the videos a surface assembled its answers from on one day.

Common Questions About Video and AI Answers

Should my company start a YouTube channel to get cited in AI answers?

Not on this evidence, and not for that reason. The ten videos cited across four category answers belong to review publishers and independent creators, and none to a company competing in the category. A brand channel publishing product content is doing the same thing its product pages do, which is describing one product, and the question that produces these answers asks for a comparison. There may be good reasons to have a channel; being cited in a category answer is not one this data supports.

Why does YouTube get cited so much in the first place?

We can tell you it does, in more categories than any other domain in our counting and in all three markets we measured, and we cannot tell you why from this data. What we can now add is that the videos doing it are comparative: rankings, tier lists, top fives and head to heads. That is the same format that wins in text, so the simplest reading is that the format is doing the work and the platform is where that format happens to live for these questions.

Who are these creators and how do I find mine?

Ask your category’s buying question, look at which videos the answer cites, and open them. That is the entire method here and it took one query per category. The channels we found are a mix of review publications like Merchant Maverick and VPNpro and individuals like Stewart Gauld and Jamie Trull, and the list for your category will be different people.

Is paying a creator to feature my product a good idea?

We have not measured it and we are not going to guess. What the data says is that the slot in the answer is held by that kind of channel, and what it does not say is what happens to a citation when the video is sponsored. Note that at least one cited video’s transcript contains an affiliate prompt, so commercial relationships already exist inside this layer, and nothing about the citation flagged that.

Can one video be cited in several countries at once?

Not in anything we have measured. Asking 3 categories in 4 markets, each in its own language, produced 12 answers that cited 25 distinct videos, and 0 of those videos appear in two markets. In all 12 cells the cited videos came back with a transcript in that market’s own language. So a video is a placement in one market, and a category you sell in four is four productions rather than one. Whether a single channel can span two markets is a different question and this did not measure it.

Does this mean video matters more than written content?

No, and the study does not compare them. In the same four answers the written sources include PCMag, Zapier, Reddit, Gartner and vendor pages, and video is one part of the source list rather than most of it. The finding is about who owns the video part, not about video winning.

What is the cheapest useful thing to do with this?

Run your own category’s buying question once, list the YouTube URLs it cites, and open each one to see whose channel it is. If they are creators, you have a list of named people who already make the format that gets cited in your market. If one of them is a competitor’s channel, you have found something we did not, and it is worth telling us about.