A few months ago, a lead came through our WhatsApp qualification flow and answered the "how did you find us" question with four words we had never seen in three years of running funnels: "ChatGPT recommended you."
No Google search. No Instagram ad. No referral. A language model, inside a private conversation we will never see, compared the options in our category and named our company. The lead arrived pre-sold, skipped the usual questions, and booked a meeting in under ten minutes.
If you run a business, this article is about making that happen for you. It covers what generative engine optimization (GEO) actually is, how AI search engines like ChatGPT, Perplexity, and Google AI Overviews decide which businesses to recommend, the exact playbook I ran across my own two companies, the popular tactics that turned out to be a waste of time, and a 30-day plan you can run without hiring anyone. I have receipts for all of it, because I did this on my own websites before writing a word about it.
Quick context on who is talking: I run Alex Digital 360, which builds WhatsApp and AI sales automation for businesses across Colombia, Mexico, and Chile, and Scala Technologies, the software agency behind it. AI search optimization started as an experiment for us in 2025. It's now a permanent line on my monthly calendar.
What actually happened to search (and why your rankings earn less every month)
Let's kill the two lazy narratives first.
Narrative one: SEO is dead. It isn't. Gartner predicted in early 2024 that traditional search volume would drop 25% by 2026, and that prediction did not survive contact with reality. Google still handles over 90% of the world's searches. If someone opens their pitch with "SEO is dead," they are selling you a course.
Narrative two: nothing has changed. That one is worse. Pew Research tracked the real browsing behavior of 900 adults and found that when Google displays an AI Overview, users click a traditional result 8% of the time, versus 15% when there is no AI answer. Ahrefs measured the same effect from the publisher side: by December 2025, having an AI Overview above your #1 ranking cut its clickthrough rate by 58%. Roughly 60% of all Google searches now end without a single click.
So the honest picture in 2026 is this: you can hold the #1 position you spent years earning and watch it produce half the clicks it used to, because an AI summary above you is answering the question. The ranking is the trophy. The click was always the money. And they are quietly separating.
Meanwhile, the other channel is compounding. ChatGPT passed roughly 900 million weekly active users by February 2026, more than double the year before, and drives about 87% of all AI referral traffic to websites. Around 77% of consumers say they used AI to help with a purchase decision in the past six months, and one in four say ChatGPT's product recommendations beat Google's. AI-sourced website traffic grew over 500% year on year in 2025, and Semrush projects that AI search visitors will overtake traditional organic visitors by 2028.
What is generative engine optimization, and how is it different from SEO?
Generative engine optimization (GEO), sometimes called answer engine optimization (AEO), AI search optimization, or LLM SEO, is the practice of making your business visible and recommendable inside AI-generated answers: ChatGPT responses, Perplexity citations, Google AI Overviews, Gemini and Claude recommendations.
The name doesn't matter much. The mechanics do, because they are genuinely different from classic SEO:
- SEO optimizes pages to rank in a list. GEO optimizes your entire brand footprint to be named in an answer.
- SEO is won with backlinks and on-page authority. GEO, per the largest study to date, is won with brand mentions across the open web, which outweigh backlinks roughly three to one.
- SEO rewards the winner with a click. GEO often rewards you with something better: the AI presents your business as the answer, so the visitor arrives already convinced.
- SEO compounds over years. GEO leaderboards are new and volatile, which means early movers gain ground absurdly fast compared to mature SEO.
The two are complementary. Everything that makes you citable by an AI also tends to be decent SEO. But the reverse is not true, and that asymmetry is where the opportunity lives right now.
The business case: AI visitors convert like referrals, not like traffic
The volume of AI search traffic is still small. What makes it worth chasing is the quality per visitor.
Semrush's study found the average AI search visitor converts at 4.4 times the rate of a traditional organic visitor. Seer Interactive's platform data goes further: ChatGPT referrals converting around 16%, Perplexity around 10%, against under 2% for Google organic.
I will give you the counterevidence too, because vendors never do. Amsive studied 54 websites and found LLM traffic converting about the same as organic. A separate ecommerce study of 973 sites found ChatGPT referrals converting worse than Google. The pattern across all these studies is consistent: AI referrals behave like warm referrals for service businesses and B2B, where the AI acts as a pre-sales consultant, and like ordinary traffic for ecommerce, where it's just another door to the store.
We sell services, and our numbers match the optimistic end. AI-referred leads complete our five-question WhatsApp qualification flow at nearly double the rate of paid-ad leads. They open conversations with questions like "how fast can you deploy," which is not a lead behavior, it's a customer behavior. If you sell services, consulting, software, healthcare, education, or anything where a recommendation carries weight, this channel deserves your attention this year, not next.
One more thing most businesses miss: this traffic is nearly invisible in your analytics. A customer asks ChatGPT for a recommendation, gets your name, then types your URL directly or Googles your brand. Your dashboard records "direct" or "branded search," and the AI's role disappears. We only see it because our qualification bot asks every lead how they found us as a required step of the flow, and people answer a WhatsApp message more honestly than any attribution survey. If you do nothing else after reading this, add that one question to your intake. It costs nothing and it will tell you whether AI is already sending you customers. That flow, incidentally, is the same system from our guide on replacing a sales team with WhatsApp automation.
How ChatGPT decides which businesses to recommend
The GEO advice industry appeared overnight, and most of it is recycled SEO folklore. So I went looking for actual evidence, and there is more of it than you'd expect: a Princeton-led study presented at KDD 2024 that tested nine optimization methods across 10,000 queries, Ahrefs' correlation analysis of 75,000 brands, Semrush's citation studies, and Profound's analysis of 680 million AI citations. Here is what they collectively say.
Your Google rank barely matters. Ahrefs found that ChatGPT cites pages ranking at position 21 or deeper in Google almost 90% of the time. The overlap between Google's top 10 and AI citations collapsed from about 75% in mid 2025 to under 40% by early 2026. The AI is not reading Google's leaderboard. It has its own, and it's still being written.
Brand mentions beat backlinks three to one. In the 75,000-brand study, the strongest predictors of AI visibility were YouTube mentions (0.737 correlation) and branded web mentions (0.664). Backlinks, the currency of two decades of SEO, scored 0.218. What other people say about you, anywhere on the open web, outweighs who links to you. Related analyses attribute over 80% of AI citations to earned, third-party media rather than brands' own content.
Listicles are the doorway. For recommendation queries like "best WhatsApp automation companies," best-of listicles account for 43.8% of ChatGPT's citations. Not your homepage. Someone's numbered comparison with your name on it.
Review platforms form the opinion but rarely get the citation. In one SaaS study, 100% of tools ChatGPT recommended had Capterra profiles and 99% had G2 profiles, yet those platforms received under 1% of visible citations. The model reads reviews to decide what it thinks of you, then quotes a blog post when explaining itself.
Statistics, quotes, and definitive language get you cited. The Princeton study found that adding statistics to a page lifted citation likelihood by 41% and quotations by 28%, with sites ranked fifth or lower gaining up to 115% visibility. Ahrefs found definitive, unhedged sentences nearly double citation odds compared to hedge-everything corporate prose.
Freshness is a real signal. Content cited by ChatGPT is on average 393 days newer than what ranks for the same query in Google. Stale evergreen content is optimized for a search engine that increasingly answers without showing it.
The GEO playbook I ran on my own companies
Everything below was run on Alex Digital 360 and Scala Technologies over roughly eight months, in order of what I now believe mattered most.
1. Get into comparison content, or publish it. If listicles drive 43.8% of recommendation citations, the move writes itself. Pitch the existing roundups in your niche, and publish honest comparison content of your own, because AI engines cite comparison pages even when a participant wrote them. Our most-cited page is exactly that: a detailed comparison of the best WhatsApp chatbots in our market. It shows up in AI answers more than anything else we've published.
2. Make your money pages quotable. An AI assembling an answer needs sentences it can lift. "WhatsApp message open rates in Latin America exceed 95%" is liftable. "Engagement tends to be strong on messaging platforms" is not. We rewrote our key pages to carry specific, sourced statistics in plain declarative sentences, following the Princeton findings. Numbers, sources, no hedging.
3. Answer the literal questions people ask. Nobody types keywords into ChatGPT. They ask full questions: how much does WhatsApp automation cost, can a bot qualify leads, does this work for a clinic. Your pages should answer those exact phrasings in two to four clean sentences each. The FAQ at the bottom of this article is not decoration; it is the tactic, applied to itself. Our article on the real cost of WhatsApp AI automation exists in the structure it does for the same reason.
4. Fix your entity consistency. Large language models build an internal picture of who you are from every mention of you across the web. If half the internet describes you as a "digital marketing agency" and the other half as a "WhatsApp automation company," the model's picture of you is blurry, and blurry entities don't get recommended. Same name, same one-line description, same service list: website, LinkedIn, Google Business Profile, directories, social bios. This costs nothing and I rank it above every technical tactic on this list.
5. Publish where the models already read. Semrush's most-cited domains study shows ChatGPT leaning on Wikipedia, with Medium, Forbes, and LinkedIn high on the list, while Perplexity leans heavily on Reddit. So we publish long-form work on Medium alongside our own blog, and every branded mention of our companies on those domains is a brand mention on a property LLMs demonstrably trust. Yes, that means writing about GEO on a top-cited domain is itself GEO. I'm at peace with the recursion.
6. Treat reviews and community as testimony. The Capterra finding reframed review platforms for me: the model reads them to form its opinion even though it rarely cites them. So keep review profiles alive and current, and show up usefully in the Reddit and YouTube conversations where your customers ask for recommendations. Not astroturfing, which platforms punish and models increasingly detect. Actual answers, in actual threads. Slow and unscalable, and per the correlation data, roughly three times more valuable than link building.
7. Keep dates visible and content current. A 393-day freshness advantage is not a rounding error. We now revisit our highest-value pages quarterly, update the numbers, and keep the dates visible.
What doesn't work: the GEO tactics the data quietly buried
This section is why this article exists. Every GEO listicle recommends the following. The evidence does not.
llms.txt does approximately nothing. The pitch: add a special llms.txt file so AI crawlers can understand your site. I did it early and watched nothing happen, and then the data arrived. Google's John Mueller stated that no AI service has said they use llms.txt, and server logs show they don't even check for it. Ahrefs looked across 137,000 domains and found 97% of llms.txt files received zero requests. Ever. It's a standard nobody adopted, kept alive by tools that generate it. Ours is still on the server, a small monument to doing what the listicle said.
Schema markup is hygiene, not a growth lever. I rebuilt structured data across both sites expecting movement and could attribute none. Citation studies have found no correlation between schema coverage and LLM citations, and the Princeton study never tested it. Keep your schema, because Google says structured data helps its own AI surfaces, and it's cheap. Just refuse politely when someone tries to sell you a "schema for ChatGPT" project.
Publishing more content doesn't move the needle. Content volume correlates with AI visibility at 0.194, which is statistical noise. Ten mediocre posts are worth less than one comparison page an AI actually cites. Publish less, make it citable.
Chasing every engine at once dilutes everything. Each engine eats differently, which brings us to the next section.
Engine by engine: where each AI actually looks
Profound's analysis of 680 million citations makes one thing clear: there is no single "AI search." Optimize for the engine your customers actually use.
- ChatGPT favors Wikipedia-adjacent authority, listicles, and comparison content, and cites Medium, Forbes, and LinkedIn heavily. Reddit is only about 11% of its citations. It drives about 87% of all AI referral traffic, which makes it the default priority for most businesses.
- Perplexity is a Reddit engine: nearly 47% of its top citations come from Reddit threads. If your customers are developers, researchers, or early adopters, community presence matters disproportionately here.
- Google AI Overviews lean on Quora and Reddit, respect Google's own structured data guidance, and are the one surface where your classic SEO and schema investments carry over most directly.
- Gemini and Claude publish less citation data, but entity consistency and third-party mentions appear to drive both, consistent with the general pattern: what the open web says about you is the input.
For our market, LATAM service businesses, ChatGPT is the overwhelming winner, so that's where we optimize first and measure monthly.
The 30-day GEO plan for a small business
You do not need an agency to start. Run this exactly as ordered.
Week 1: audit. Open ChatGPT and Perplexity and ask what a customer would ask: "best [your category] in [your city]," "who should I hire for [your service] in [your country]." Note who gets named. Those names are your real competitors in AI search, and the list probably won't match Google's. Save the answers; they're your baseline.
Week 2: fix your entity. One name, one description, one service list, everywhere your business appears online. Update your Google Business Profile, directories, and review platforms. The models read reviews to form opinions even though they don't cite them.
Week 3: make one page quotable. Take your main service page and give it specific numbers, sourced claims, definitive sentences, and an FAQ answering the five questions customers actually ask you on the phone. Not more pages. Better sentences.
Week 4: earn one third-party mention. One listicle inclusion, one podcast appearance, one industry roundup, one genuinely useful Reddit answer in a thread where your customers ask for recommendations. Third-party mentions outweigh your own content roughly three to one, so one earned mention beats a month of blogging.
Then repeat the week-one audit every month and watch whether your name starts appearing. This moves faster than SEO ever did, because you're not fighting twenty years of accumulated authority. The leaderboard is new, and most of your competitors don't know it exists.
The LATAM and Spanish-language window
This part is for our readers in Colombia, Mexico, Chile, and the rest of Latin America, and it's the closest thing to free money in this whole article.
The Spanish-language side of AI search is nearly uncontested. Ask ChatGPT in Spanish for the best providers in almost any LATAM niche and the answers are visibly thinner than their English equivalents, drawn from generic, Spain-centric content. There is almost no citable Spanish comparison content for Colombian, Mexican, or Chilean markets. The listicle ecosystem that feeds AI recommendations in English barely exists in Spanish.
That means the playbook above, executed in Spanish for a LATAM market, competes against approximately nobody. Whoever builds the citable comparison pages, the statistics-dense guides, and the consistent entity footprint for a LATAM niche in the next year becomes the default recommendation, because the models will have almost nothing else to cite. We've watched this work in our own market, where WhatsApp is how recommendations become conversations: the AI names a business, and the customer's next move is a WhatsApp message. If that message lands in a qualification flow that responds in three seconds, the loop from AI recommendation to booked meeting closes without a human touching it.
This window will not stay open. LATAM GEO agency listicles are already starting to appear. Early is exactly when this is cheap.
Where AI search is heading
Semrush puts the crossover, AI search visitors passing organic visitors, around 2028. I think purchase intent crosses over earlier: the valuable slice of discovery, the "who should I hire, what should I buy" questions, is moving into AI conversations while Google keeps navigational leftovers.
The step after that is already visible: agents that don't just recommend but act. When a customer's AI assistant can shortlist three vendors and message them directly, the winners will be the businesses the model already trusts, whose information is consistent and quotable everywhere, and who can receive that conversation instantly on the channel it arrives on. In Latin America, that channel is WhatsApp, which is why we treat AI visibility and WhatsApp automation as one system, not two projects.
Rank was the moat for twenty years. The new moat is being the name the machine says when nobody is watching.
Frequently asked questions about GEO and AI search
How do I get my business to show up in ChatGPT recommendations?
Focus on the signals the data supports: get your business mentioned in third-party comparison articles and listicles, keep your name, description, and services identical everywhere online, maintain active review profiles, and publish statistics-dense, quotable content on your site and on domains LLMs cite heavily, like Medium and LinkedIn. Brand mentions across the web outweigh backlinks roughly three to one for AI visibility.
What is generative engine optimization (GEO)?
GEO is the practice of optimizing your business's web presence so AI engines like ChatGPT, Perplexity, Gemini, and Google AI Overviews mention and recommend it in their answers. It overlaps with SEO but runs on different signals: brand mentions, comparison content, quotable statistics, entity consistency, and freshness matter more than backlinks or Google rankings.
Is GEO different from AEO?
The terms are used almost interchangeably. Answer engine optimization (AEO) usually emphasizes structuring content to answer questions directly, while generative engine optimization (GEO) covers the full practice of earning visibility in AI-generated answers. In practice, the tactics are the same.
Does SEO still matter in 2026?
Yes. Google still handles over 90% of searches, and AI Overviews draw partly on Google's index. But clicks from traditional rankings are declining, with AI Overviews cutting top-position clickthrough by up to 58%. The practical answer: keep your SEO, and add GEO on top, because the two share most of their groundwork.
How does ChatGPT decide what to recommend?
ChatGPT synthesizes an opinion from its training data and live web retrieval, weighting third-party sources heavily: best-of listicles account for 43.8% of its citations for recommendation queries, and it cites pages ranked 21 or deeper in Google almost 90% of the time. Review platforms like G2 and Capterra shape its opinion even though it rarely cites them directly.
Does llms.txt help my site get cited by AI?
The evidence says no. Google's John Mueller confirmed no major AI service uses llms.txt, and an Ahrefs study across 137,000 domains found 97% of llms.txt files received zero requests. It doesn't hurt, but treat it as optional, not as a strategy.
Does schema markup improve AI visibility?
Partially. Google says structured data helps its own AI surfaces, so keep it for AI Overviews. But citation studies have found no correlation between schema coverage and ChatGPT or Perplexity citations. Schema is cheap hygiene, not a growth lever.
How long does GEO take to show results?
Faster than SEO. AI citation patterns refresh quickly and favor content roughly 400 days newer than Google's rankings, and the competitive field is thin. Businesses running a focused effort commonly see themselves appearing in AI answers within one to three months, especially in low-competition niches and non-English markets.
How do I measure whether AI is sending me customers?
Dashboards mostly can't see it, because AI-referred visitors arrive as direct traffic or branded searches. The reliable method is asking. Add "how did you find us" as a required question in your intake or qualification flow and track the answers. Monthly, re-run your category questions in ChatGPT and Perplexity and record who gets named.
Does GEO work for small and local businesses?
Arguably better than for anyone else. AI answers for local and niche queries are built from thin source material, so a single well-built comparison page, a consistent entity footprint, and a handful of earned mentions can make a small business the default recommendation in its niche, particularly in Spanish-language and LATAM markets where citable content barely exists yet.
The short version
Customers are increasingly asking AI who to hire, and the AI's answer is built from signals most businesses aren't managing: third-party mentions, comparison content, quotable statistics, consistent entity data, and freshness. The tactics that work are boring and free. The tactics being sold hardest, llms.txt files and schema miracles, have no evidence behind them. And the Spanish-language LATAM side of this is an open field for roughly one more year.
We build the receiving end of this pipeline for a living: WhatsApp and AI sales systems that turn an AI recommendation into a qualified, booked meeting in minutes. If you want to see what that looks like in practice, our case studies show the full loop, and if you want it running on your business, you know where the button is.
