The strongest measured signal isn't your website. It's YouTube.
Across 75,000 brands, the thing most closely associated with appearing in AI answers was how often a brand is mentioned in video. Backlinks barely registered. Here is what that does and does not mean.
Ahrefs, 75,000 brands, 27 May 2026 · correlational · August 19, 2026 · 5 min read
In May 2026 Ahrefs published the largest study anyone has run on this question. They took 75,000 brands, measured whether each one appeared in AI answers across ChatGPT, Google AI Mode and AI Overviews, and then looked at which other things went along with that presence, out of everything they could measure.
The ranking of signals is not what the industry has been selling.
In the right-hand column, 1 would mean “these two always go together” and 0 would mean “they have nothing to do with each other”. None of these numbers proves a cause. They only say how closely the two things travel together.
How closely it travels with showing up in AI answers
How often the brand is named inside videos 0.74
How many people watched those videos 0.72
How often it is named elsewhere on the web 0.66 - 0.71
How often the name appears inside other people's links 0.63
How many people search for the brand by name 0.47
How authoritative the site is considered 0.27 - 0.33
How many pages the site has 0.19
How many links the site receives very weak
Read the top and the bottom together, because that’s the finding.
The top of the list is entirely about being talked about elsewhere. Video mentions, web mentions, branded anchors, branded search. Four measures of the same underlying thing: how often your name appears in material you did not write.
The bottom of the list is your website. How many pages you have, how authoritative your site is considered, how many links you have collected. And links, the currency this trade has run on for twenty-five years, came out so weak that the company reporting it was Ahrefs, which makes its living selling a tool for links.
An earlier study by the same team found the same thing from another angle: how many people search for your name travelled with AI visibility, and the links you collect did not. The citation data agrees. In a June 2026 check of 3,543 citations across ChatGPT and Google AI Overviews, the ones pointing at a company’s own website numbered two. Two.
The caveat, before the recommendation
This is a correlation, and the direction of causation is genuinely unresolved.
A brand that is mentioned in a thousand YouTube videos is, generally, a brand that people already know. Fame produces mentions. Mentions may also produce AI visibility. Both can be true, and this study cannot separate them. It is entirely possible that a good part of that top number just means “famous brands are famous.”
So the wrong reading is: buy video mentions, appear in AI answers. Nobody has demonstrated that. An experiment would need brands randomly assigned to a mention campaign, measured before and after against a control group, and that study does not exist. When someone eventually runs it, the effect will almost certainly be smaller than the table makes it look. That is exactly what happened with the descriptive code behind pages, and what moved wasn’t the code tells that story.
Here is what makes the finding useful anyway.
Even if the numbers are inflated, the order holds. Whatever share of them is fame rather than mechanism applies to every row in the table equally. Fame also produces links, pages and authority, and those still came bottom. The signals that measure other people describing you beat the signals that measure you describing yourself, consistently, across 75,000 brands. That ordering survives the caveat.
And it agrees with something we can watch directly. The systems that fetch pages for AI go and quote other people’s pages. That is not a number to interpret, it is what the citation records show. The Ahrefs ranking and the citation split are two independent measurements pointing at the same structural fact: these systems build answers about you out of material written by other people.
What that means practically
Not “make YouTube videos.” Something more specific and considerably cheaper.
The signal is being mentioned, not publishing. Ahrefs notes that being named often across many small videos beat one appearance in a big one. Ten local channels, review videos, industry round-ups, somebody else’s walkthrough that names you: that is what the number counted. A polished company channel with four videos and no mentions anywhere else is not the same thing.
Being nameable matters more than being visible. For a mention to help, your name has to be attached to a clear description of what you do and who for. A paragraph, not a page has the passage shape. The question your customer asked has the reason: you are not competing for the question your customer asked, you are competing for the six questions the system made out of it, and a mention that answers one of those is worth more than a mention that just says your name.
The cheap version, for a business with no money for video. What sits at the top of that table is not YouTube. It is your name turning up somewhere, and video is only one way that happens. Trade press, local coverage, supplier and partner pages, industry directories that are actually read, podcast appearances, someone else’s comparison post, transcripts. Each of those is a page with your name on it that you did not write. And when a video exists, what gets said in it becomes written text that programs can read, which is a large part of why video mentions count for so much.
Do the reverse search first. Before commissioning anything, find out what already exists. Search your business name and see what pages carry it. In most small-business categories the answer is: a handful, most of them wrong or five years old. Correcting a stale directory entry and getting one accurate regional page published is a week of work, and it moves the exact signal that sits at the top of this table.
The thing to stop paying for
If you are currently paying somebody to go and get you links, on the theory that authority pours over into AI answers, this is the study that should change that line in your budget. Across 75,000 cases links came out weak, and the company publishing it sells tools for links. You will not find a more disinterested result in this field.
Move it to being described accurately, by other people, in more places.
