Why AI recommends your competitor and not you

A woman in Guadalajara opens ChatGPT on a Tuesday night and types: "which real estate agency should I use to sell an apartment in Providencia?". In three seconds she has an answer with three names, one line on why each, and no desire to keep searching. She decides from that list.

Your agency isn't on the list. Not because you're worse. Because the AI never considered you.

And here's the uncomfortable part: to build that answer, the AI barely looked at your website. According to AirOps' 2026 State of AI Search, roughly 85% of a brand's mentions in AI answers come from external sources —directories, reviews, forums, articles— not from the brand's own site.¹ More than that: brands with a strong presence off their own site are 6.5× more likely to win AI visibility.¹ The digital brochure you fussed over barely counts; what others say about you does.

From ranking to being selected

Google trained you to think in rankings: ten blue links, ordered, and you chose. AI doesn't do that. It pulls fragments from many sources and synthesizes them into a single answer —and in that step, it pre-selects for you. It went from ordering to choosing.

The difference isn't cosmetic. In a ranking, #8 still exists: someone can scroll down and find you. In a selection of three, #8 doesn't exist. There's no page two. The AI decided you weren't relevant, and that decision is invisible —to the user and to you.

So the question stops being "what position do I rank in?" and becomes a harder one: "did the AI even consider me a candidate?".

The five signals AI weighs

When an engine decides who goes into those three names, it doesn't flip a coin. It synthesizes signals. AirOps' analysis identifies five that weigh more than the rest.¹ They're worth taking one at a time, because each maps to something a business owner recognizes.

1. That your information can be extracted. AI reads to cite, not to admire your design. A restaurant in Guadalajara that posts its menu as a pretty image on Instagram is illegible to a model; the same menu in text —dishes, prices, hours— is citable material. AirOps measured that pages with well-structured data (for example, three tables) get up to 25.7% more citations.¹ It's not aesthetics: it's whether the AI can copy and paste what's yours.

2. Specificity. "Quality care" tells a model nothing. "Root canals, invisible braces, same-day emergencies, 9 to 8" does. A dental clinic in Monterrey that lists its services precisely hands the AI phrases it can use verbatim. Princeton's GEO research found that adding verifiable data and cited sources lifts visibility in generative engines by up to 40%.² The vague is ignored; the specific gets cited.

3. That others talk about you. This is the one that stings, because you don't fully control it. AI trusts what third parties say more than what you say. A taquería in Puebla mentioned in a forum, a local article and thirty text reviews outweighs one with a flawless site and silence around it. It's no accident that roughly 48% of AI citations come from communities like Reddit and YouTube¹: it's third-party text, open and attributable. AI looks for consensus, not self-promotion.

4. Consistency across platforms. If your real estate agency in Querétaro declares one coverage area on Google, another on its website and a different phone number on Facebook, the AI doesn't know which to believe —and when in doubt, it doesn't recommend you. Models reward the coherent entity: same name, same area, same data, everywhere. Consistency is a trust signal that's cheap to give and expensive to ignore.

5. Freshness. A spa in Tulum that updated its services and hours this month tells the AI it's alive and operating. A listing frozen in 2023 suggests otherwise. Engines favor recent signals; fresh content competes better than content that hasn't moved in years.¹

Why your competitor does

Now put the five together. Your competitor probably isn't bigger or better. Their information is in extractable text, they describe what they do precisely, they accumulate third-party mentions, they say the same thing everywhere and keep it fresh. You maybe check one or two. The AI, which neither knows nor likes you, adds up signals and picks whoever clears more boxes.

One detail throws many owners off: ranking well on Google doesn't guarantee appearing in AI. Recommendations for local businesses lean on entity signals —that you exist as a recognizable, consistent business— more than on a site's authority. For services, AI also factors in geographic relevance and the proximity implied in the question. That's why a business that dominates Google sometimes doesn't appear in ChatGPT, and a modest one on Google does. Different games, different rules.

What to do with this

The takeaway isn't "produce more content." It's understanding that each of those five signals is something concrete you can measure and fix. But you can't fix what you can't see.

That's where Plexa comes in. A Plexa listing is your business turned into something AI can read and cite: extractable structure, specific data, a consistent entity that says the same thing everywhere. And the visibility report does what no Google dashboard does: it tells you in what fraction of AI answers you appear today, against whom, and which of these signals you're losing on.

It's not magic or smoke. It's measuring before optimizing. Start by knowing where you stand: if AI already recommends your competitor, the first step is to see exactly why.

Footnotes

1. AirOps — 2026 State of AI Search (2026). ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶

2. Princeton — GEO: Generative Engine Optimization (Aggarwal et al.). ↩