Transcript
Intro
Here's a result that should change how you spend your off-page budget. A study of seventy five thousand brands checked what lines up with AI assistants bringing a brand up. The strongest factor wasn't backlinks. It was how often the brand gets mentioned on YouTube.
So do links still matter for AI answers, or is the whole game now about being talked about? The honest answer sits in between. Today we find out where, run a ten prompt test on a brand, and finish with local businesses, where mentions have a name of their own: citations.
First, a quick picture of the problem. Your link report says you had a great quarter. Then a customer asks an AI assistant for the best site in your category, and you're not in the answer. Both things can be true at once, because the two systems don't count the same things.
What the studies measured
Back to that study. It came out in December 2025 and covered seventy five thousand established brands, all with a Domain Rating above forty. It checked which signals rise together with being named in ChatGPT, Google's AI Mode and AI Overviews.
Here are the numbers. YouTube mentions came out on top at about 0.74. Brand mentions across the web landed between 0.66 and 0.71, depending on the platform. Domain Rating, the link metric everybody chases, sat between 0.27 and 0.33. Raw backlink counts were described as very weak.
Now the one warning label this episode needs. Big brands get mentioned more, searched more and linked more, all at once, so a correlation like this can come from brand size alone. Small brands weren't even in the sample. Read it as a direction, not a formula. Nobody has shown that a YouTube mention adds a set amount of AI visibility.
A second study points the same way. In January 2025, an agency ran ten thousand real search questions from finance and software through GPT-4o and watched which brands it named. Ranking on page one of Google lined up with being mentioned at about 0.65. The backlink effect was, in their words, weak or even neutral.
Why would mentions matter at all? Google's own July 2026 guide to AI search drops a plain hint. Its generative AI features can surface what people say about products and services in blogs, videos and forum threads. An assistant answering a question about your category reads pages about your category. If nobody on those pages talks about you, there's nothing to repeat.
Put it together and a sensible reading appears. Links still help you rank, and ranking still helps you get named. The AI layer just adds a second currency: how often credible people talk about you. Links are one way to earn it. They're not the only way, and they're not the strongest thing anyone has measured.
Earned vs paid
So where do AI assistants find what people say about you? A May 2026 study looked at more than twenty five million links cited by ChatGPT, Claude and Gemini, across seventeen industries. Eighty four percent of those citations came from earned media.
And paid or advertorial content? Zero point three percent. Journalism alone was twenty seven percent. The earned share has stayed between eighty two and eighty nine percent in every edition since July 2025, so this isn't a one-month blip. It's a vendor study, so the source line says so.
That has a practical consequence for anyone with a link budget. A sponsored post on a news site can still be fine for traffic, and Google allows paid links when they're labelled sponsored or nofollow. As a route into AI answers, though, it barely registers. Coverage someone chose to write about you is what gets read.
Let's be clear about our stance, because this is a link building series. Links aren't dead. In our field work, carefully earned and carefully placed links still move rankings in many industries. What the AI data adds is that real coverage pays twice: once as a link for Google, and once as a mention an assistant can read.
In practice, the digital PR from the last episode is also your AI mention strategy. A data story about film trailers that lands on a few film news sites gives ReelRadar links and places where an assistant reads its name next to the topic. Judge those campaigns on both, and you'll stop shrugging at coverage that came without a link.
Used vs named
Now a twist most dashboards miss. An AI assistant can use your page as a source and never say your name. A June 2026 study of almost four thousand brand appearances across four AI engines tracked two things separately. Was the page cited as a link? And was the brand actually named in the answer?
Sixty one point seven percent were ghost citations. The page was linked as a source, and the brand never appeared in the text. Your page did the work, someone else took the bow. Only thirteen point two percent were cited and named. The other twenty five point one percent were the reverse: named, with no link at all.
And the engines behave very differently. ChatGPT cited a source in eighty seven percent of appearances but named the brand only about one time in five. Gemini did almost the opposite. It named brands in eighty three point seven percent of appearances and linked them only twenty one point four percent of the time.
So when you track AI visibility, keep two columns, never one. Mentioned means your name is in the answer. Cited means your page is in the sources. And look at that named-but-not-cited group: a brand can be recommended without a single link to its own site.
Claim: unlinked mentions are worthless. A link builder's reflex is to chase every unlinked mention and ask for a link. Asking is fine, dismissing the mention isn't. A quarter of brand appearances in that study were names with no citation, and earned coverage is exactly where those names come from. Verdict: placebo.
The ten prompt test
Now the receipt you can build yourself. Take ten prompts a real customer would type. For ReelRadar, our fictional film trailer and where to watch site, that means prompts like: where can I watch the trailer for a new release, and which site shows UK release dates. Mix short category prompts with full situations, because wording changes results a lot.
Run each prompt in four assistants, for example ChatGPT, Google's AI Mode, Gemini and Claude. For every answer, log three things. Was ReelRadar named, was it cited, and which third-party pages did the work: the reviews, lists and threads the assistant leaned on. That last column tells you where to spend.
Write the prompts in customer language, not marketing language. Steal the words people use in support emails, trailer comments and forum threads. A prompt your own team writes will flatter you, because it'll contain the phrases you already rank for. You're looking for where you're missing, not for a nice screenshot for the board.
Then turn the log into a plan. Group the third-party pages by type. If assistants keep leaning on best-of lists you are missing from, that is your target list for the next quarter. If they lean on YouTube breakdowns, a few honest, useful videos may do more than another round of guest posts. The data tells you which door to knock on.
Here's how that plays out for ReelRadar. Say the log keeps showing the same film magazine roundup of trailer sites, and ReelRadar isn't on it. That one page is now worth more than a dozen random links, because assistants keep reading it. The job is to deserve a spot on it: better data, a reason for the writer to look.
And run it more than once. AI answers change from run to run, so one screenshot proves nothing. Repeat the same ten prompts every week and watch the share of answers that name you over a month. We measure that properly in Part 3. Here, the useful output is the list of third-party pages, because those are the doors worth knocking on.
The second half of the receipt is the brand description check. Write one sentence that says what ReelRadar is. Then line it up against the about page, the social profiles, directory listings and the press release boilerplate. If they disagree, an assistant reading five sources meets five slightly different businesses.
Fix the mismatches first. It is the cheapest work in this episode, and it removes a reason for an assistant to describe you wrongly. One sentence, used word for word, everywhere you control.
Fake mentions
Which brings us to the shortcut people reach for next. If mentions matter, why not just make some? A few fake accounts, some forum posts praising the brand, a batch of glowing reviews. It sounds cheap. It usually ends up expensive.
Google answered this directly in its July 2026 guide to AI search: seeking inauthentic mentions across the web isn't as helpful as it might seem. Its core systems focus on quality, other systems block spam, and the AI features depend on both. The same guide lists chasing inauthentic mentions among the tactics to ignore.
This is different from the advanced link methods we covered earlier in Part 2. Those can work in the right hands, because a link passes value through a graph a careful operator understands. A fake mention has to fool the actual people in the thread. When they spot it, the brand wears it. The risk sits on your name, not on a domain you can let go.
Local citations
Now local businesses, where mentions have their own name. In local SEO, a citation is a listing of your business name, address and phone number on another site: directories, maps, data aggregators. Meet the Corner Screen, a fictional neighbourhood cinema that partners with ReelRadar for showtimes.
Google's own help page says local results are mainly based on relevance, distance and prominence. Prominence comes from things like how many websites link to your business and how many reviews you have. And Google's profile guidelines ask that your business name match the real-world name, used consistently on your storefront, website and stationery.
Reviews belong in the same picture. The same Google page says more reviews and positive ratings can help a business's local ranking. For the Corner Screen, that means asking real customers after a screening, replying to reviews, and never buying them. A fake review is just a fake mention wearing a star rating.
The best current read on local factors is the big 2026 survey of forty seven local SEO experts, published in November 2025. For map results, the weight given to citations slipped by about half a percentage point, continuing a slow decline. But in a new column for AI search, three of the top five factors were citation factors. Their summary: in AI SEO, mentions are the new link.
That's expert opinion, not measurement, so weigh it as informed experience. It does match the logic. An assistant answering which cinema near me shows classic films has to pull business facts from somewhere, and listings and best-of lists are where those facts live.
Claim: submit your business to three hundred directories. Google's spam policy lists low-quality directory or bookmark site links as link spam, and using automated services to create links is on the same list. Bulk directory packages tend to be both at once. Verdict: placebo, with a spam label attached.
What works instead for the Corner Screen is short and boring, which is how you know it's real. Get the main listings exactly right: the Business Profile, Apple Maps, Bing Places and the handful of directories that matter in its country. Then earn local mentions from real relationships: the local paper's what's on page, a film festival sponsor list, a university film society.
The audit takes an afternoon. Search for the Corner Screen's name and phone number, and list every place it appears with its name, address, phone, website and opening hours. Look for the old phone number from before the refit, the old name the cinema traded under, a listing that still sends people to a closed second screen. Each of those is a fact an assistant might repeat.
Wrap up
Here's the episode on one screen. Mentions line up with AI visibility more strongly than link metrics do, earned media dominates what assistants cite, and the shortcuts fail for the usual reason: they're fake.
The AI mention tracker sheet and the listing consistency checklist are free on the episode page at irongoo.com. Run the ten prompts this week, fix your brand description, and then decide where your next mention should come from. Next episode, we measure a link campaign and show you how we prove the wins.
