Why Social Buzz Is Not AI Visibility
My wife has been very into Korean skin care lately, and we both use AI quite a bit. So, when she sent me a WWD article on the “Buzziest Skin Care Brands Online in Europe,” I thought I’d run a test. If I used the headline as a prompt: “What are the buzziest skin care brands online in Europe?”, would AI’s response reproduce the article?
I had just read McKinsey’s report that across France, Germany, and the UK, 84% of people surveyed use AI tools in everyday life, and 38% said they use AI to research products and services or decide what to purchase. So, the test idea wasn’t entirely out of the blue.
The article my wife sent me was based on data from CreatorIQ, a platform that measures Earned Media Value across social platforms. The list was specific. Ten brands, ranked in order.
Erborian — $25.5M
Garnier — $23.2M
La Roche-Posay — $22.1M
CeraVe — $18M
Nivea — $17.1M
Medicube — $16.8M
Kiehl’s — $16.2M
L’Oréal Paris — $15.4M (tie)
Rhode — $15.4M (tie)
Caudalie — $10.9M
The WWD article was paywalled at the time. During our test, a free syndicated version also appeared on Yahoo Shopping, but automated access to that page was blocked as well. So what would AI do with the question?
What the models found — and what they didn’t
One model retrieved the article and reproduced the list accurately. All ten brands in the same order.
A second found the WWD headline and teaser copy ahead of the paywall and filled in the rest of its response from a skincare blog and what the model already knew.
The third didn’t find the WWD article at all. It pulled from different sources — a beauty index, a TikTok ranking platform, a lifestyle site — and produced an entirely different list. A few brands that had earned a valuable spot on the original “buzziest brands” list either dropped lower on AI’s version or were replaced by others.
One model reproduced the same list as the article. The other two illustrated a problem for brands: inclusion on a fact-based time-sensitive ranking that is behind a paywall, or other barriers, needs another step if the brands on the list want AI to recognize them for it.
Says who?
We asked each model to audit its own responses. In this second round of prompts, almost every claim was described by the model that made it as either weakly supported or unsupported. The brand groupings and explanations each model offered had very few external sources behind them. One model’s ranking rested on a snippet it couldn’t fully read, something it only acknowledged when asked directly the second time around, not in its original response.
Yet the models all answered with the same confident tone regardless of how weak their evidence was. A skin care consumer or journalist reading those responses had no way to know which parts were anchored by quality sources and which were backfilled by what was in memory.
The brands didn’t help themselves
All ten brands on the list sell from their own websites or refer visitors to storefronts. Between the article’s March 17 publication and our March 23 test, a search of each brand’s website, press page, and news section showed no mention of the WWD ranking. Though beyond WWD, the ranking did appear in two other locations: Yahoo Shopping and a Greek lifestyle site that republished it in Greek.
This matters because in our test, the models that found the WWD article used it as their primary source, while those that couldn’t, due to the paywall or server access issues -- details that only emerged in our follow-up audit -- filled the gap with lower-authority alternatives. The brands that made the list should proactively promote it, lest AI miss it.
What about social media?
These consumer-facing, search-enabled models don’t work the way a social analytics platform does. They don’t normally scan TikTok, count Instagram mentions, or survey what’s trending on YouTube in real time. In our test, the models appeared to rely primarily on indexed pages — articles, blogs, published reports — not the underlying social activity those articles describe.
A brand posting fifty times a week probably does not, from that activity alone, become more visible to these models. Social content can influence AI visibility, but indirectly. High engagement drives EMV, EMV gets reported in trade press. Trade press gets indexed. AI reads the trade press. Every step is a point where the benefit to a brand can be lost -- behind a paywall, blocked from automated access, or simply never written about.
“Flood the zone” is not an AI strategy
Some brands are being told to produce more social content so AI recognizes them as current and relevant. That’s not wrong for building social engagement. But to answer, “What are the buzziest skin care brands online in Europe?”, the models favored accessible articles with specific claims, structured data on owned properties, and press coverage with data in a form AI could reach.
Most brand teams haven’t examined what AI says about them, where that information comes from, and what’s being missed. The brands on the WWD list earned a ranking that can impact their sales. They may not know that getting AI to recommend them to consumers researching their category takes some additional effort. If you work with brands, try the test on your own category.
This post first appeared in the Brand AI Report, my newsletter on how AI describes and recommends brands.