Being mentioned on YouTube correlates with AI visibility better than links do

Ahrefs states the caveat, and the announcement of the study drops it

A YouTube mention, in this study, is any time a brand name crops up in a YouTube video title, transcript or description. Not a view, not a subscriber, not a watch minute. The definition is the finding, and it is the first thing to fall off in the retelling. The sample was domains with a Domain Rating above 40, each taken with its highest volume keyword at 800 monthly searches or more, which is how 75,000 brands were found. The coefficient is Spearman, a rank correlation.

So the strongest correlate of appearing in an AI answer is a text field on a video platform. YouTube mention impressions, the same mentions weighted by how many people saw them, came in slightly lower at 0.717, and Ahrefs draws the conclusion out loud: brands do not appear to be at a huge disadvantage if they are mentioned in low view videos, so long as they are mentioned widely. Frequency beats reach.

Then read the announcement. The Business Wire release that carried this into the trade press, dated 27 May 2026 on the syndicated copy, is headlined that YouTube mentions are the strongest signal of AI visibility across 75,000 brands. Its body does use the word correlated, twice. What it carries nowhere is a coefficient, or any version of the causation line: the release says YouTube is the strongest signal and never says how strong. The Next Web, covering the same study, kept both the method and the disclaimer, so this is not a complaint about journalism. It is a note about which document a headline writer reads.

The caveat survives in the study and does not survive in the announcement of the study.

The comparison with backlinks splices two studies of the same brand set

The December 2025 study covers three platforms and says only this about links: very weak correlations between number of backlinks, URL rating and brand mentions across all AI systems. The numbers behind that sentence sit inside a chart image. The figure everyone quotes, 0.218 for number of backlinks, comes from the earlier study of 26 May 2025, which used the same 75,000 brand construction but measured AI Overviews only: branded web mentions 0.664, branded anchors 0.527, branded search volume 0.392, Domain Rating 0.326, referring domains 0.295, backlinks 0.218.

Putting 0.737 next to 0.218 means taking the YouTube factors from one study and the link factors from another, of the same brands, on one of the three platforms. That is what the chart below does, and it is defensible as long as it is not done quietly. The ranking survives the disclosure: content volume, the number of pages on a site, came in at roughly 0.194 in the December study, and the top three correlates in the AI Overviews study are all off site factors. The two do not agree to the decimal either: branded web mentions read 0.664 in May and 0.656 in December for the same platform.

One more disclosure deserves the attention it did not get. The branded web mentions Ahrefs studied also included mentions from youtube.com, where the brand appeared in the video title, so the second ranked factor is not fully independent of the first. That is what gets lost when a bar chart is screenshotted out of the post explaining it.

The same vendor added JSON-LD to 1,885 pages and citations did not move

On 11 May 2026 Ahrefs published the other kind of study. They tracked 1,885 web pages that added JSON-LD between August 2025 and March 2026, matched each against three control pages from different domains with similar pre-period citation levels that never added schema, 4,000 controls in total, and measured citation change across AI Overviews, AI Mode and ChatGPT. Adding schema produced no major uplift on any platform: AI Overview citations on treated pages fell 4.6 per cent relative to control, AI Mode rose 2.4 per cent and ChatGPT rose 2.2 per cent, the last two indistinguishable from zero. They will not attribute the decline to schema either, because both groups were already falling before the markup went on. One boundary on the result: every page in the sample already had at least 100 AI Overview citations in February 2025, so it measures what schema does for a page that is already cited.

Ahrefs had analysed 6 million URLs and found AI cited pages almost three times more likely to carry JSON-LD, the kind of gap they describe as getting shared in LinkedIn carousels and conference slides as proof that schema is an AI visibility lever. Their own answer to why 53 per cent of cited pages run schema: the sites that add structured data also invest in technical SEO, publish authoritative content, build links, maintain their pages and rank well in ordinary search. The control group is the whole argument. Raw before and after growth for AI Mode was 43 per cent, and once the control URLs were subtracted it shrank to 2.4.

Google's own AI features page, last updated 10 December 2025, says the same thing: you do not need to create new machine readable files, AI text files or markup to appear in these features, and there is no special schema.org structured data that you need to add. A searchVIU test on 30 October 2025 looked at what the retrieval step actually reads, across ChatGPT, Claude, Gemini, Perplexity and Google AI Mode. During direct fetch every system extracted only visible HTML content, and JSON-LD, hidden Microdata and hidden RDFa were all ignored.

One measurement moved a press release and the other moved nothing, and they came out of the same tool.

YouTube is 5.6 per cent of cited AI Overview URLs, not the quarter being quoted

Ahrefs put a denominator on YouTube's real footprint in March 2026: 863,000 keyword SERPs and 4 million AI Overview URLs. Of the cited pages that did not rank in Google's top 100 for the same keyword, 18.2 per cent were YouTube URLs, and those were 5.6 per cent of every AI Overview URL cited. YouTube is the most cited domain in AI Overviews and has grown 34 per cent over six months, and it is still one cited URL in eighteen.

Otterly.ai, published on 2 March 2026 and measuring more than 100 million AI citation instances over 30 days, gives its share in two pieces: 5.54 per cent of citations came from social and video domains, and YouTube was 31.8 per cent of that subset. Multiply them and YouTube is about 1.76 per cent of everything Otterly saw, roughly a third of the Ahrefs figure and counted across six assistants rather than one Google feature. Both are a long way below the headline shares of 23 and 29.5 per cent that circulate with no corpus size attached. A percentage without a denominator is not a measurement of anything.

Otterly found views, likes and subscribers show near zero correlation with citation frequency, roughly -0.03, which is the same result as Ahrefs finding mention count above mention impressions. Its sharpest detail is narrower: 94 per cent of the YouTube citations it saw went to long form videos rather than Shorts, and every timestamped citation it recorded came from Google, 73 per cent in AI Overviews and 27 per cent in AI Mode, none anywhere else. Google names the long form split itself in its Search Console documentation, where the suggested comparison for a YouTube property is URLs containing /watch against URLs containing /shorts/.

The transcript is the part of a video an answer engine can read

Put the two findings side by side and the common term is text: the strongest correlate is a title, a transcript and a description on a video page. The intervention that failed was hidden text, ignored at fetch time by every system searchVIU tested. Google's list of fundamentals for AI features asks for important content to be available in textual form, and its only mention of structured data asks that the markup match the visible text on the page.

Which is why publishing video has stopped being purely a brand decision. Google now reports YouTube, TikTok, Instagram and X performance inside Search Console as platform properties alongside a website, documented as of 29 July 2026. A channel is being treated as a search surface by the search engine.

None of this promises a mention. A Spearman coefficient of 0.737 over 75,000 brands cannot tell anyone that a video will be cited, and the earlier study found 26 per cent of the brands it sampled had zero AI Overview mentions at all. What the pair of studies does say is where mentions of brands are observed to live, and where an hour of effort has already been measured doing nothing. Publish the video with a transcript somebody corrected, and spend the schema afternoon on the rich results schema was built for.

What to do about it

1. Write the brand name into the video title, the description and the spoken script, and publish a transcript you corrected by hand, because that text is what the 0.737 was measured on.
2. Stop funding JSON-LD as an AI visibility tactic for pages that already get cited. That is the population Ahrefs tested, 1,885 of them against 4,000 controls, and the markup moved citations by nothing anyone can attribute to it.
3. Ask any AI visibility statistic for its denominator and its control group before you repeat it, and quote the study rather than the press release about the study.