Image: Charles Floate Twitter
A few weeks ago, SEO veteran Charles Floate created a fake awards program called “The Charles Floate Annual SEO Awards.” No committee, no ceremony, no winners, no history.
Just a name he made up and a decision to see how quickly he could make Google believe it was real. You can read the full case study here.
By day three, it ranked number one organically for “SEO awards” and Google’s AI Overview described it as “a prominent, free-to-enter program for the global SEO community.”
Prominent. An award that did not exist a week earlier.
The experiment is uncomfortable reading. But it confirms something I have been applying in client work for the past two years, and it is worth unpacking for anyone trying to understand how AI search actually decides what to surface.
The machine does not know what is real. It only knows what is corroborated.
Google’s AI Overview did not rank that fake award because it verified the organisation existed. It cannot do that. What it did was look at the signals around the entity and measure whether they were consistent with the pattern of a real one.
A dedicated website. An about page. Named judges. Categories. Dates. A reference-style page describing it neutrally. A mention on an established personal brand site with existing entity authority.
To a system that determines truth through corroboration, four consistent sources describing the same entity become consensus. And consensus is what gets cited.
Charles called it “pattern matching” rather than verification. That is exactly right.
What this means for legitimate SEO and GEO
I want to be clear: I am not advocating for manufacturing fake entities. The risk is real and when Google patches the obvious signals, the penalties will be severe.
But the underlying mechanism Charles exposed is the same one that makes legitimate GEO work. And understanding it changes how you build an organic presence.
Here is what I have observed across client work:
A page that screams “best plumber in Fort Lauderdale” will not get cited by an AI engine. But a plumber with 200 Google reviews, mentions in three local publications, a consistent NAP across 40 directories, and a named owner with a LinkedIn profile and a local business association membership? That entity has the correct shape. The machine reads it as real because it is real, and corroborated consistently.
The same principle applies in B2B. A SaaS company with one well-optimised service page will not appear in ChatGPT answers about their category. A SaaS company with a thought leadership cluster, analyst mentions, a G2 profile, customer case studies on third party sites, and a founder who publishes on LinkedIn? That entity has built corroboration at scale. Legitimately.
The three things I now prioritise for AI visibility
First, third party mentions. Not self-published content. What do other sources say about this business? Reviews, press mentions, directory listings, industry citations. AI engines trust what others say about you far more than what you say about yourself.
Second, entity structure. Named people, consistent business information, clear associations between the brand and its category. The machine needs to be able to build a picture of who this entity is and what it does from multiple independent sources.
Third, topical clusters. A single great page is not enough for AI citation. A site that covers a topic comprehensively, with interlinked supporting content, signals topical authority. LLMs prefer sources that own a topic, not sources that have one good page about it.
The honest takeaway
Charles proved the machine is gameable. That is not a surprise to anyone who has been in SEO long enough.
But it also proved something more useful: the signals that create AI visibility are the same signals that create genuine authority. Reviews, mentions, entity associations, topical depth. The difference between what Charles built and what I build for clients is that mine is real.
The machine cannot tell what is real. But it rewards the same patterns either way.
