Table of Contents

• GEO, SEO and AI search: a summary
• What GEO is
• The paper that started it all
• GEO, AEO, LLMO or just SEO: the acronym debate
• My conclusion: GEO is expanded SEO
• How generative engines pick their sources
• What actually seems to matter
• The click drop: what it means for your business
• Schema, llms.txt and other myths
• How to measure whether you're showing up
• An opportunity for Spanish-language content
• Continue with the practical guide
• Related sources
• Editorial note

GEO, SEO and AI search: a summary

Aspect	Traditional SEO	GEO	My take
Goal	Show up in organic results	Be found, understood and cited by AI	Complementary goals, not substitutes
Outcome	Link, ranking and click	Mention, citation, recommendation or link	A brand can gain visibility even with fewer clicks
Dominant factor	Relevance, technical SEO and links	Brand mentions off your site and domain authority	Off-site reputation matters more than most guides admit
Measurement	Rankings, impressions, CTR and conversions	Citations, share of voice and referred traffic	You need to review real answers, not just a "score"

Before going into detail, here's the short version: the content typically taught about GEO gets the concept right — structure your content so AI can discover, trust and cite it — but it almost always falls short on one key point. The factor that correlates most strongly with getting cited in 2025-2026 isn't the structure inside your page, it's the mentions of your brand outside of it. I break this down with data below.

What GEO is

GEO stands for Generative Engine Optimization: the work of increasing the odds that a brand or piece of content gets found, understood and cited by ChatGPT, Google AI Overviews, Gemini, Perplexity and other generative systems.

The paper that started it all

The term "GEO" wasn't coined in a marketing blog — it came from an academic paper: GEO: Generative Engine Optimization (Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande), presented at KDD 2024 and also available as a preprint on arXiv since November 2023.

The authors simulated a generative engine in two stages: Google would retrieve the top five sources for a query, and GPT-3.5-turbo would write an answer citing them. They tested nine different tactics across a bank of roughly 10,000 queries (GEO-bench). The best performers were adding citations and sources, adding statistics, and explicitly citing the origin of the information — with visibility gains that, in the paper, exceed 40% on some queries. Keyword stuffing, on the other hand, performed worse than doing nothing.

It's an interesting finding, but it has limits that almost never get mentioned when the study is cited: it was done with GPT-3.5 (already outdated), on only five sources per query, and the authors themselves allowed the prompts to "invent" hypothetical statistics and citations to maximize the effect. Two of the three winning tactics essentially boiled down to adding new content to the page — so part of the improvement could simply come from adding information that wasn't there before, not from "GEO optimization" in any strict sense. I treat it as an academic starting point, not a definitive recipe for 2026.

GEO, AEO, LLMO or just SEO: the acronym debate

There's no consensus on what to call any of this, and the disagreement isn't purely semantic. A survey by Aleyda Solís (SEOFOMO, 2025) found that around a third of professionals prefer "AI search optimization" outright; others use AEO or LLMO; the SEO tooling industry — Ahrefs, Semrush — settled on GEO.

There are some fairly skeptical voices on the proliferation of acronyms:

• John Mueller (Google), in August 2025: the more urgency and new acronyms some people push, the more likely they're just "spamming and scamming."
• Rand Fishkin (SparkToro) argues we shouldn't abandon "SEO" and warns that any tool promising an "AI ranking position" is unreliable, because generative answers are probabilistic.
• Aleyda Solís holds that roughly 95% of AI optimization is still being led by SEO teams: she treats it as an evolution, not a break.
• Google, in its official documentation: "from Google Search's perspective, optimizing for AI-powered generative search is optimizing the search experience, and is therefore still SEO."

My conclusion: GEO is expanded SEO

I don't think GEO replaces SEO. Most of its recommendations — useful content, authorship, structure, evidence and authority — already belonged to good organic positioning.

The real difference is that we're no longer always competing for a link in a list of ten results. We're also competing to be one of the sources an AI chooses to build an answer from — something that can happen without the user ever clicking anything.

Google explains that its generative features still depend on how it crawls, indexes and understands pages in its official AI optimization guide.

How generative engines pick their sources

Each system works differently, and it's worth not treating them as one and the same:

• ChatGPT (with search) has no index of its own: it uses Bing's index to retrieve results and then cites from them. An analysis by Seer Interactive covering 80 million Semrush queries found that 87% of ChatGPT's citations match results that were already ranking in Bing's top results. The practical implication is blunt: if your site isn't indexed in Bing, you effectively don't exist for ChatGPT.
• Google AI Overviews and AI Mode use RAG ("grounding") over Google's own index, leaning on its usual ranking systems, and Google confirms they can use a "query fan-out" technique — firing off several related searches on subtopics before building the answer — though its own documentation uses the hedge "can use," not "always uses."
• Perplexity retrieves in real time and cites pages with structured data more often than ChatGPT does.

In all three cases, the process is probabilistic, not a fixed position: Ahrefs documented that AI Overviews change frequently from one day to the next, and various studies show that the same question asked repeatedly can return a different set of cited brands most of the time. Any exact "weighting" of factors you see cited on a blog (for example, "domain authority 40%, quality 35%") is unconfirmed reverse-engineering — treat it with skepticism.

What actually seems to matter

Clarity helps: answer first, elaborate after. An analysis by Kevin Indig covering millions of ChatGPT citations found that 44% of citations come from the first 30% of a page's content — a fairly consistent pattern.

Evidence matters too: cite sources, separate facts from opinions, and add your own examples, in line with what the Princeton paper found.

But the point most guides underrate is this: mentions off your site matter more than the structure inside it. An Ahrefs study covering 75,000 brands found that brand mentions across the web correlate at 0.664 with visibility in AI Overviews, versus just 0.218 for traditional backlinks — a threefold difference. YouTube, Reddit and Wikipedia are among the most-cited domains by generative engines; YouTube in particular is, according to several analyses, the single most-cited domain in AI Overviews.

On top of that, ranking in Google's top 10 no longer guarantees you'll be cited in an AI answer the way it used to: the overlap between organic rankings and AI Overview citations had been dropping sharply through 2025-2026 according to various Ahrefs analyses. At the full-domain level the correlation stays strong, but at the individual-URL level, it's weakening.

For a creator brand, this is a key idea: an article can turn into a video, a transcript, a newsletter, a collaboration and a conversation. Authority is built as an ecosystem, not as an isolated page.

The click drop: what it means for your business

When an AI Overview shows up, click behavior changes noticeably. Pew Research (2025), analyzing nearly 69,000 real searches, found that users click on any link only 8% of the time when an AI Overview is present, versus 15% when it isn't; and only about 1% click specifically on one of the sources cited within the AI summary itself.

The flip side is that the traffic that does arrive from AI platforms converts better: various Ahrefs and Semrush analyses put that traffic's conversion rate several times higher than that of normal organic traffic, even though the absolute volume remains small compared to traditional search.

The practical takeaway: don't expect GEO to spike your sessions. Think of it as brand building and high-intent lead capture, not as a traffic-volume lever.

Schema, llms.txt and other myths

Structured data is still useful for describing articles, people, organizations and videos. Official documentation lives at schema.org and in its getting started guide.

However, there's no magic schema that guarantees you'll show up in a generative answer. Google says so explicitly in its documentation: structured data isn't required for AI-powered generative search, and there's no special schema.org type you need to add. In fact, Google retired the HowTo rich result in 2023, and the FAQPage one was restricted that same year before disappearing entirely later on — the markup is still valid on schema.org, but it no longer produces any visible rich result in Google. If a course or guide is pitching you FAQ/HowTo schema as a "trick" to appear in AI, it's outdated.

I also wouldn't put llms.txt among the first priorities for a small site. Google, through John Mueller and Gary Illyes, has confirmed multiple times that it doesn't use it and doesn't plan to support it, comparing it to the old keywords meta tag: a file almost nobody on the search engine side reads. No major system (OpenAI, Anthropic, Perplexity) documents it as a citation signal. Its real use today is closer to serving as documentation for coding tools (Cursor, Copilot) than as a search visibility lever.

What's still worth keeping: Article, Organization, Person, Product, Review, VideoObject and ImageObject, because they help build your brand's entity in the knowledge graph and still produce rich results Google hasn't retired.

How to measure whether you're showing up

There's already a category of tools dedicated to this — Profound, Peec AI, Otterly.ai, Ahrefs Brand Radar, Semrush AI Toolkit, among others — with pricing that ranges from roughly $25-30/month at the entry level up to enterprise contracts. They all measure variations of the same thing: share of voice in AI answers, citation rate, brand sentiment, and which sources appear alongside yours.

Before paying for a tool, you can do this manually: pick 10 to 20 real questions related to your brand or niche, ask them in ChatGPT, Gemini and Perplexity, and log whether you show up, alongside which competitors, and whether there's a link. Be wary of any tool that only gives you an "AI position" number without showing you the full answer: generative responses are variable, and a single run doesn't tell you much.

An opportunity for Spanish-language content

Generative search still has uneven coverage in Spanish and across Latin America: several industry sources agree that specialized Spanish-language content is underrepresented among the sources AI engines cite, despite Spanish being spoken by more than 500 million people. I wouldn't treat this as a hard, quantified figure — it's more of an industry consensus — but it's a reasonable hypothesis to experiment with while the window stays open.

It's worth adding regional context, local examples, currency, country and natural language, because models adjust their answers based on country and language variant. Firsthand experience can differentiate more than a generic translation of English-language content.

If you're looking for Spanish-speaking voices to keep digging into this: Aleyda Solís (Orainti, SEOFOMO newsletter), Fernando Maciá (Human Level, with his "semantic consensus" concept) and Lino Uruñuela (Mecagoenlos.com) are longtime voices in Spanish-language technical SEO who are also covering this topic.

Continue with the practical guide

If you want to move from research to implementation, read How to Implement GEO: A Practical Guide for Creators.

You can also check out Frequently Asked Questions About GEO, SEO and AI Search.

Related sources

• Original GEO paper on arXiv
• GEO: Generative Engine Optimization (KDD 2024, DOI)
• Google's official guide on AI optimization
• Google's guide on AI-generated content
• Official schema.org documentation
• Pew Research study on AI results

Editorial note

This field moves fast. Many of the recent studies cited here are correlational and come from companies that sell SEO or AI-visibility tools (Ahrefs, Semrush, Profound), which doesn't invalidate them but does call for context: they aren't peer-reviewed papers, and the exact percentages vary between studies. Each platform's internal source-selection mechanisms are black boxes; nobody outside those companies knows the real weights. It's worth checking the date, methodology and sources before turning a percentage into a promise.