Obility Editorial · · 4 min read
The YouTube correlation measures fame as much as video
Ask whether YouTube helps a brand appear in AI answers and most replies point to one study. Ahrefs looked at 75,000 brands and found that the number of times a brand is named on YouTube, in a video’s title, description or transcript, had the strongest correlation with how often ChatGPT, AI Mode and AI Overviews mentioned it, a Spearman coefficient of about 0.737. Many posts turn that into advice to start a company channel.
Ahrefs itself warns that correlation is not causation, and the design shows why the warning matters here. It counts mentions of a brand anywhere on YouTube, mostly in other people’s videos, and brands that creators talk about are brands people already talk about everywhere else. The same study found branded web mentions correlated almost as strongly, between 0.66 and 0.71. A brand’s own uploads are a small slice of what was measured, so the figure says little about what a new channel would do.
Outside engines read the page around the video, not the audio
For an answer engine to use a video, it has to get text out of it. The text YouTube serves openly is the watch page: the title, the description and the channel name. We fetched a watch page the way a crawler would. The description was in the HTML. The captions were not; the page only pointed to them at an address under /api/timedtext.
YouTube’s robots.txt file gives every search crawler the same rules, and it disallows /api/ along with the internal paths the page uses to load its other features. A crawler that follows the file can read what the description says but not what the presenter says. A product demo in which the founder explains exactly who the product is for, posted with a one-line description and a list of social links, leaves an outside engine almost nothing to quote.
Google has a route the other engines lack
Google owns YouTube and does not have to crawl it the way an outside company does. Its Gemini API documentation shows one consequence: a developer can pass a public YouTube link and the model processes the video itself, a feature Google marks as a preview. That is a developer feature, not a description of how AI Overviews or AI Mode choose sources, and Google has not published how its search features use spoken video content. It is still reasonable to expect what is said in a video to count for more in Google’s answers than in ChatGPT’s, Perplexity’s or Claude’s.
Two other routes exist, and you control neither. Models can learn from transcripts that ended up in their training data, and an assistant fetching a link a user pasted may not follow the same rules as a search crawler. Neither helps when a buyer asks a category question without naming you.
Make the words around the video carry the claims
Treat the description as the copy outside engines will read. Open with what the product is and who it suits, in plain words. State the specific claims the video makes, such as the integrations shown, the team size it fits and the problem it solves, and add chapter titles that name each topic. A description that reads as a short article does more work than one that reads as a list of links.
Then give the spoken content a home you control. Put the video on a page of your own site with the full transcript, edited for reading, or a written version of the same argument. That page is open to every engine your site already admits, can be cited directly, and keeps the substance if the video is moved or taken down.
Mentions in other people’s videos are where the Ahrefs signal actually sits. A reviewer’s title and description that name you and your category are readable text on a page you do not own, which is the kind of independent mention answers draw on. Ahrefs found that the count of mentions correlated slightly more than their views did (0.737 against 0.717), which points toward several accurate mentions rather than one large video.
Check whether video shows up in your answers at all
Before funding video for AI visibility, look at what your category’s answers cite. Obility’s Answer Engine Insights keeps the answer, the brands it names and the sources it cites with each observation, so you can see question by question whether youtube.com appears among the sources, and on which engines. The engines covered for your programme are confirmed at the demo. If video never appears for your buying questions, it is not the lever there, whatever the aggregate studies say.
A simple test separates audio from text. Pick a fact that is said aloud in one of your videos and written nowhere else, ask each engine a question that needs it, and note which engines can answer. Then add the fact to the description or the transcript page and repeat the question over the following weeks. Answers vary from run to run, so compare repeated runs rather than single replies.
This is about AI answers only. A good video still persuades the buyers who watch it, and a transcript page does not replace it. The page makes sure the argument in the video can be read by the engines that cannot watch.