GEO Optimization Guide — 전체 시리즈

  1. 1. What Is GEO - AI Citation Strategy Beyond SEO ← 현재 글
  2. 2. Each AI Cites Different Sources
  3. 3. On-Site GEO Technical Architecture - From Product DB to JSON-LD
  4. 4. Off-Site GEO - How to Win Over AI That Ignores Your Official Site
  5. 5. AEO - Why Coding Agents Read Documentation Differently

Ranking on Google’s First Page Felt Like Enough

You worked hard on SEO and landed on Google’s first page. Organic traffic naturally increased. So far, a familiar scenario.

But lately, the way people search has changed. They type “recommend a budget laptop” into ChatGPT or look up “family-friendly hotels in Seoul” on Perplexity. Google AI Overview now drops answers right above the search results.

The Shifting Search Paradigm

Clicks are disappearing. Because AI is answering on behalf of the user.

Of the URLs cited by AI, only 9% rank in Google’s top 10 (Ahrefs). High SEO rankings do not guarantee AI citations. A separate optimization layer is needed.

That layer is GEO (Generative Engine Optimization).

What Is GEO

GEO is about making AI search engines like ChatGPT, Perplexity, Gemini, and Google AI Overview cite our content in their answers. It is the work of restructuring data itself so that AI can read it easily.

SEO was about getting people to click. GEO is about getting AI to name us as a source.

AspectSEOGEO
GoalDrive clicks (Traffic)Get cited in AI answers (Citation)
Trust criteriaKeyword density, backlink countIdentifiability, structured data
Recognized byHumans + search engine botsGenerative AI models
Core techniquesMeta tags, content optimizationJSON-LD, Schema.org, FAQ structuring
KPIRanking, CTRMention rate, citation accuracy

One thing to be clear about: adopting GEO does not mean abandoning SEO. A solid SEO foundation is what makes GEO work. GEO is built on top of it.

Why Now

The data shows the tide has already turned.

ChatGPT WAU has surpassed 800 million (OpenAI). South Korea is the world’s second-largest paid subscription market, with one in three economically active people using AI. Gartner projects that traditional search volume will decline by 25% by 2026 (Gartner). According to a Capgemini report, more than two-thirds of consumers actually purchase products recommended by AI (Capgemini). Over half of Google searches end as zero-click results, and this ratio will only climb higher once AI Overview is layered in.

The conversion side is even more compelling. The purchase conversion rate from AI search traffic is 14.2% – 5 times higher than traditional Google search (GrackerAI). Revenue per AI-referred visit is also more than 2.5 times higher (Adobe).

Traffic is declining, yet conversion rates for AI-curated results are actually higher. What drives revenue is no longer how widely you are exposed, but whether AI selects you.

The Three Principles of GEO

When applying GEO, certain questions keep coming up. Can AI distinguish this product from others? Does it understand the purpose and context? And once it reads the content, is the structure there for it to cite the source?

Three GEO Principles: Identity, Context, Citability

Identity - Identifiability

AI must be able to clearly distinguish a product or service.

International standard identifiers like GS1 GTIN/GLN are key. For AI to recognize “ChocoStick Original” and “ChocoStick Almond” as separate products, each needs its own unique GTIN. If they are bundled under a single representative code, AI cannot tell them apart.

Context - Contextual Connectivity

AI must be able to understand the purpose, relationships, and positioning of a product.

Category hierarchies, variant relationships (flavor/size/color), brand-product-SKU structures – these contexts need to be structured so AI can match the right product to a query like “running shoes for men in their 20s.”

Citability - Citation Readiness

The content must be structured so that AI can read it and attribute the source.

JSON-LD, FAQ Schema, and robots.txt configuration fall under this principle. No matter how good the data is, if AI crawlers cannot access it or if the structure is hard to parse, AI will simply skip it.

Here is the summary table:

PrincipleKey QuestionCore TechnologyVerification Criteria
IdentityCan AI distinguish this from others?GS1 GTIN/GLNIdentifier registration, variant differentiation
ContextDoes AI understand the purpose and relationships?Categories, knowledge graphMetadata quality, cross-channel consistency
CitabilityCan AI read this and cite the source?JSON-LD, FAQ, robots.txtStructured data validity, crawler access permission

Invisible GEO vs Visible GEO

Implementation splits into two approaches.

Invisible GEO is JSON-LD inside the <head> tag. It is invisible to users but directly parsed by AI and search engines. It is the most powerful method for boosting AI citation rates. However, if your site is an SPA, a transition to SSR (Server-Side Rendering) must come first.

<!-- Invisible GEO: JSON-LD inside <head> -->
<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "ChocoStick Original",
  "gtin13": "8801234567890",
  "brand": { "@type": "Brand", "name": "K Foods" },
  "offers": {
    "@type": "Offer",
    "price": 1500,
    "priceCurrency": "KRW",
    "availability": "https://schema.org/InStock"
  }
}
</script>

Visible GEO is HTML content inside <body>. FAQ pages, detailed product descriptions, nutrition tables. Both humans and AI read it. The technical barrier is low, so you can start right away.

ItemInvisible GEOVisible GEO
Location<head> JSON-LD<body> HTML
SEO impactHighModerate
AI citation rateHighHigh
Implementation difficultyHigh (requires SSR)Low
User experienceNone (machine-only)Directly visible

In practice, both are used. JSON-LD is added for machine parsing, while HTML is laid out for both humans and AI to read.

AI Already Answers Well – the Problem Is the Source

“Recommend a family-trip hotel.” I threw this query simultaneously at Genspark, Perplexity, and ChatGPT. All three returned similar answers – pool information, room prices, breakfast details. Answer quality is already sufficient.

The problem is the source. Genspark directly cites Schema data from official websites. Perplexity scrapes from blogs and travel platforms. ChatGPT references official sites but loses precision without structured data. The same hotel, yet each AI shows a different price.

Schema does not guarantee official citations. What it does is make official sites easy for AI to parse, increasing the probability that AI chooses the official source over a blog. That is why On-Site GEO matters.

What the Research Says

A research team from Princeton and Georgia Tech analyzed 10,000 queries, and the results are quite clear:

  • Content with explicit sources: AI visibility +40%
  • Content with statistics: +30%
  • Content with repeated keyword stuffing: actually -10%

Keyword repetition, which worked in SEO, actually hurts in GEO. AI does not care how many times the same word appears – it looks at how systematically organized and trustworthy the information is.

Empirical evidence on structured data is also accumulating. Brands cited in AI Overview see a 35% higher organic search CTR, and paid ad CTR nearly doubles (Seer Interactive). Adding structured data increases the probability of appearing in AI Overview by 36% (GrackerAI). Sites with complete Schema have an 80% chance of appearing in ChatGPT, compared to 20% with only basic Schema (Search Engine Land).

Content freshness cannot be overlooked either. More than three-quarters of pages highly cited by Perplexity were updated within the past month. Pages untouched for over three months get pushed down.

On-Site GEO and Off-Site GEO

GEO broadly splits into two domains.

AspectOn-Site GEOOff-Site GEO
DefinitionMaking your own site readable and citable by AIExposing your brand on external sites that AI references
Core techniquesJSON-LD, Schema.org, SSR, robots.txt, FAQReddit, Wikipedia, news outlets, communities
ResponsibilityDev team / technical organizationMarketing / PR / brand strategy

This series focuses on On-Site GEO – the domain where developers can apply changes directly through code.