Ask ChatGPT "what recruitment agencies do you recommend?" and it will introduce several services. But are those companies actually the ones the market chooses?
I don't think that's necessarily the case. In reality, there are companies that sell well but aren't chosen by AI — and companies that are chosen by AI but don't actually sell.
Companies chosen by AI and companies chosen by humans share some similarities, but their evaluation criteria aren't identical. This article explores those differences.
What AI and Humans Are Each Looking At
When evaluating a brand, AI and humans look at very different things.
| What AI looks at |
What humans look at |
| Volume of public information |
Direct experience |
| External mentions |
Word of mouth |
| Consistency of information |
Sense of trust |
| FAQ, case studies, articles |
Support quality |
| Ease of explanation |
Price and terms |
AI has never actually used a service. So the only thing it can evaluate is publicly available information. Humans, on the other hand, factor in everything — how the sales team communicates, the price, and post-implementation support.
To dig a little deeper: through training on vast amounts of web text, AI builds associations between brand names and specific categories and contexts. In the world of AI, this is called an Entity. Brands recognized as Entities are more likely to surface as candidates when AI generates a response; brands that aren't recognized simply don't come up. Whether or not a brand is recognized as an Entity is the starting point for whether AI chooses it at all.
This is fundamentally different from how humans evaluate. Humans listen to sales pitches, experience demos, negotiate prices, and live through post-implementation support before making a judgment. AI has none of that process. This asymmetry is what creates the gap between companies AI chooses and companies humans choose.
Companies That Sell Well But Aren't Chosen by AI
Well-known within the industry. High customer satisfaction. But AI doesn't recommend them — these companies exist.
They have a website. They have a track record. But look closer and the content is thin — brand imagery over substance. Case studies are listed as "a major manufacturer" with no specifics. There are no detailed comparison articles or FAQs. The FAQs live in the sales team's heads, not on the site. Their reputation circulates through word of mouth in the industry, but it doesn't exist as indexed text on the web.
The issue isn't that the information doesn't exist — it's that it isn't published in a form AI can cite. AI can only reference information that exists as structured, publicly accessible text. No matter how excellent a service is, if the information doesn't exist in a form AI can read, it won't make the list of candidates.
GEO research firm Xyle has stated: "AI doesn't recommend brands with the best products. It recommends brands with the best content about their products." I don't fully agree. But on the point that AI evaluates public information, I think it gets at something essential.
In the 2000s, people said "companies without websites can't be found." I feel like the same thing is starting to happen with AI now.
But this time, having a website isn't enough. We're entering an era where even if you exist, you won't be found unless your information is organized in a way AI can understand.
Companies Chosen by AI That Don't Sell
The reverse also exists. Frequently recommended by AI — but actual sales figures and customer satisfaction aren't particularly high.
AI tends to favor companies that have FAQs, case studies, comparison articles, and glossaries — companies that are easy to understand. Companies with strong content publishing can be recommended by AI even without necessarily having strong products.
It's also worth considering how reliable AI recommendations actually are. According to SparkToro research, the probability of getting exactly the same recommendation list twice in 100 queries is under 1%. AI recommendations are a "probability," not a "correct answer" — which is why tracking them through a metric called SOM (Share of Model) is more appropriate. If a company appears in 40 out of 100 queries, that means it's a candidate 40% of the time — which is separate from any judgment that the company is "good."
Furthermore, between being recommended by AI and that translating into a purchase, there's always a human judgment. Even after AI recommends a company, the actual choice comes down to the sales interaction, price negotiation, and demo quality. AI recommendation is simply the "making the consideration list" stage — nothing more.
That said, I think this case is less a GEO problem and more a marketing problem overall. Building a state where AI chooses you and having value that humans choose you for are not inherently in conflict.
GEO Is Not About Changing the Evaluation
There's something I want to clarify here.
The purpose of GEO is not to trick AI into recommending you. It's to create a state where the value you actually have can be correctly understood by AI. Concretely, that means building out case studies, comparison content, and FAQs, strengthening trust signals based on E-E-A-T, and getting your brand recognized as the right Entity by AI — all of this is "work to communicate value you already have."
Reducing the number of cases where "a good company isn't being chosen" — that's what I believe GEO is for.
AI-chosen companies and human-chosen companies may never perfectly overlap. But aiming to be "chosen by both AI and humans" is possible. That requires both the ability to communicate value through content, and the actual value to deliver. Neither alone is sufficient.
Put another way: even if you polish your content and start getting AI recommendations, it won't last if your actual service quality doesn't hold up. Being chosen by AI is not the goal — it's the entry point.
Summary
- AI and humans use different criteria to evaluate brands — AI evaluates public information, humans evaluate direct experience
- Companies that sell well but aren't chosen by AI have information that isn't published in a form AI can cite
- Companies chosen by AI that don't sell exist, but that's a marketing problem, not a GEO problem
- AI recommendations are a probability — being recommended and actually being chosen are different things
- GEO is not about tricking AI — it's about making your actual value legible to AI
- Reducing cases where "a good company isn't being chosen" is the role of GEO
Related column: For our thinking on what GEO strategy is really about, see The Core of GEO Strategy, As I See It.
Related article: On the relationship between AI citation and sales, see Getting Cited by AI Doesn't Increase Sales.