Brand mentions track AI visibility better than backlinks

Across a study of roughly 75,000 brands, branded web mentions correlated with AI visibility at 0.664 and YouTube mentions at 0.737. Backlinks came in at 0.218. If you are still budgeting link acquisition as the primary authority lever, the ratio of your spend to your results is upside down.

Before anyone reorganises around this, the caveat has to come first, because it is doing real work.

Correlation is not the mechanism

Big, well-known brands have lots of mentions, lots of YouTube coverage, lots of links and lots of AI visibility. All four move together because all four are downstream of being genuinely known. A correlation of 0.737 tells you the two things travel together. It does not tell you that manufacturing mentions produces visibility.

Google's own May 2026 guidance says the opposite, in writing: manufacturing brand mentions across the web is not as helpful. That is not a contradiction of the data. It is the resolution of it.

Mentions are a proxy for being known. Fake the proxy and you get nothing. Become known and the proxy follows.

Which is a less convenient finding than buy mentions, and considerably more useful, because it tells you where the budget should go.

Why YouTube scores highest

The 0.737 figure for YouTube is the most interesting number in the set, and the most actionable.

Video transcripts are text. They are indexed, they are retrievable, and they contain your brand name spoken in context by someone who is not you. A podcast episode where a specialist says your company name while explaining a problem is a different kind of evidence than a directory listing carrying the same string.

There is a second reason. Video coverage is harder to fake at scale than written mentions. Nobody spins up 400 YouTube appearances. So the signal stays relatively clean, which is exactly why it correlates well.

Practically, that means the highest-yield authority work for a medium or large business right now is getting your actual people into other people's video and audio. Industry podcasts, conference sessions that get recorded and published, guest appearances, panel recordings, customer webinars hosted by partners.

That work is slow, it does not scale linearly, and it cannot be outsourced to a tool. It is also the thing your market is least likely to be doing, because it requires a person with something to say.

What a mention needs to look like

Not all mentions carry the same weight, and the difference is legible.

Useful: your brand named alongside the specific problem you solve, in a source that covers your category, by someone with a reason to name you. A line saying we had a compliance reporting problem across four sites and used this firm to consolidate it is worth more than a hundred directory entries.

Useless: a logo in a supplier list. A name in a sponsor footer. A listing on an aggregator page that nobody reads. These are strings, not context.

The distinction matters when you brief PR. A placement that gets your name into a publication without getting it near the problem you solve does very little for retrieval, because retrieval is matching on context.

Practical work, in order

1. 1.Audit your current mention profile. Search your brand name with site: exclusions for your own domains and see what actually exists. Most companies discover their mentions are 80 percent directory noise and 20 percent substance.
2. 2.Fix the unlinked mentions you already have. Where somebody has described your work without naming you, or named you without context, ask for the correction. This is the cheapest gain available and takes an afternoon.
3. 3.Build the video and audio pipeline. Pick six to ten shows, channels or events that your buyers actually consume. Not the biggest, the most relevant. Pitch your practitioner, not your marketing lead. Aim for one appearance a month, sustained for a year.
4. 4.Publish something that gets referenced. Original data is the most reliable mention generator that exists, because journalists and podcasters need numbers and will name the source. One properly conducted annual survey of your own market produces more contextual mentions than a year of outreach.
5. 5.Track mentions as a metric, alongside links. Whatever monitoring tool you already run, split it into mentions with context and mentions without, and report only the first number.

What to stop

Stop buying link placements on sites nobody in your industry reads. At a correlation of 0.218 they are not worthless, but they are the weakest lever you are funding and usually the most expensive per unit.

Stop the guest-post pipeline that exists to place a link. If the piece would not be worth publishing without the link, the mention it generates carries no context.

Stop measuring PR by placement count. Count the placements where your brand appears within the same paragraph as the problem you solve. That number is usually a fraction of the reported one, and it is the only one that tracks with anything.

Where this approach breaks

If you sell into a market with almost no media surface, no podcasts, no active trade press, no conference circuit, this strategy has nowhere to run. Some industrial and specialist B2B categories are genuinely like this. In those markets the substitutes are customer-published case studies, standards bodies, and technical documentation that other people cite. Slower again, but it is what exists.

And if your product genuinely is undifferentiated, no mention strategy fixes that. People name the companies that did something notable. The uncomfortable version of this finding is that the authority problem is often a product problem wearing a marketing budget.

Being talked about is the strategy. Everything else is a way of measuring whether it worked.