GEO vs SEO: The Three Gates Before an LLM Cites Your Site

Introduction

Search is changing from a list of links to a written answer. When someone asks ChatGPT, Claude, Gemini, or Perplexity for a recommendation, the brands named in that answer win attention that used to go to the top search results. This note sorts out the terms, checks which claims are verified and which are not, and turns the discussion into a practical checklist. By the end, you will know where GEO fits relative to SEO, why JavaScript rendering matters more for AI crawlers than for Google, and how to break the vague question “Does AI recommend us?” into specific checks you can run.

One-Line Summary

GEO (Generative Engine Optimization) extends SEO rather than replacing it. LLMs choose sources on top of the same crawl, index, and ranking workflow, so a site has to pass three gates in order to appear in AI answers: bot accessibility (technical SEO), topical authority (content SEO), and mentions, sources, and citations (GEO measurement).

Core Diagram

Key Data

Terminology

Term	Definition	Optimization target
SEO	Search Engine Optimization: ranking high in keyword search results from search engines such as Google and Naver (South Korea’s largest search portal)	Crawl → index → ranking
GEO	Generative Engine Optimization: getting a brand shown inside answers from LLMs such as GPT, Claude, Gemini, and Perplexity	Source selection and citation by LLMs
AEO	Answer Engine Optimization: appearing in AI summaries at the top of search results, such as Google AI Overviews	Entry of cited sources into AI summaries

• The academic origin of the term GEO is Aggarwal et al., “GEO: Generative Engine Optimization” (arXiv 2311.09735, KDD 2024, Princeton and IIT Delhi). In controlled experiments, content optimization (adding citations, statistics, and sources) was first shown to improve visibility inside generative engines by up to about 40%.

Verified fact: AI crawlers do not execute JavaScript

• A joint study by Vercel and MERJ (December 2024, analyzing more than 500 million GPTBot fetches) found zero evidence of JavaScript execution by GPTBot, ClaudeBot, or PerplexityBot. GPTBot downloads JavaScript files with an 11.5% probability but does not run them. Googlebot is the only crawler that renders JavaScript, using a two-stage indexing process with headless Chrome.
• As a result, content on client-side rendered (CSR) sites is invisible to LLM crawlers. Product descriptions, FAQs, and comparison tables loaded from an API in React, Vue, or Angular do not exist from a GEO perspective.
• The standard response is to move to server-side rendering (SSR) or static site generation (SSG). The interviewee also described a workaround used in commerce: placing text in HTML and CSS behind an image-heavy product detail page so bots can read it. This is an observation the interviewee made in the field and has not been independently verified.
• Self-check: disable JavaScript in your browser and navigate your own site. That is roughly what an AI crawler sees.

Unsettled claim: the correlation between Google top rankings and LLM visibility is weakening

• The interviewee claimed that about 60% of AI Overview citations come from Google’s top 10 results. This matches early research, but the most recent figures have moved substantially.
  • Ahrefs: the top-10 citation share fell from 76% as of July 2025 to 38% in a later study. The remainder of citations came from positions 11–100 (about 31%) and from outside the top 100 (about 31%).
  • Originality.AI: 52%. BrightEdge: 54% of citations came from organic ranking (October 2025).
• The cause pointed to is Google’s query fan-out: a question is split into several sub-queries, and the system cites pages that perform well across that cluster. The link between “ranking on page one” and AI citation is therefore loosening.
• Conclusion: the direction “good SEO carries over to GEO” still holds, but the era in which entering the top 10 guaranteed AI citations is ending. Coverage of the whole topic cluster matters more.

Technical SEO checklist (based on the interviewee’s live diagnostic criteria)

• Meta information: title, description, schema markup (JSON-LD), favicon, and site name settings
• robots.txt: explicit allow or block declarations per bot (there is also a view that it can be used to optimize unnecessary bot traffic)
• sitemap.xml: a map that tells bots about URL structure, freshness, and priority
• Heading hierarchy (H1/H2/H3 as the large, medium, and small section structure) and image alt text. The keyword to expose is your own brand name, not the name of a partner’s logo.
• URL path clustering (a directory structure organized by topic)
• Rendering structure: whether the site uses CSR, SSR, or Suspense-SSR, and what share of information bots actually receive
• Diagnostic tools: the Chrome extension “SEO Meta in 1 Click,” and searching in guest mode to exclude personalization

GEO measurement framework: MSC (the interviewee’s framework)

Stage	Definition	Measurement
Mention	The brand is named within the sentence of an LLM answer	Whether the brand appears in the sentence, and its recommendation rank against competitors
Source	Your domain is used as material in building the answer	Whether your domain appears in the source list
Citation	A URL is attached to the end of a sentence, which happens when the model judges the domain to have high trust or authority	Whether a citation appears

• Each engine (GPT, Gemini, Perplexity, Claude) prefers different citation domains. Based on those differences, the interviewee plans earned media strategy, such as targeting Wikipedia.

Content strategy (summary of the interview)

• Mass AI publishing without a strategy does not build topical authority. A site with “one article defining AX and the rest insurance content” loses trust with people, Google, and GPT alike.
• Search journey map: understanding (what it means) → exploring (examples) → comparing → recommending → deciding. The questions at each stage become the table of contents for content.
• Natural-language questions in Google’s People Also Ask (PAA) boxes are direct clues for content topics.
• Keyword ownership strategy: for highly competitive keywords, claim the primary and secondary related keywords first, a “land-grab” approach.
• Elements LLMs seem to prefer (the interviewee’s claim, which matches the direction of GEO research): statistics and numbers, citations and sources, neutral comparison content, and TLDR and comparison-table structures.
• Resource allocation: SEO and performance marketing drive customer acquisition, while CRM drives retention. SEO takes 1–6 months to show results, so start it early, and run performance campaigns in parallel to add traffic for short-term validation.

Insights

1. GEO is built on SEO. LLMs have little incentive to build an entirely new ranking algorithm, so they reuse much of the crawl, index, and ranking output of search engines. The starting point for GEO is therefore not a mysterious new technique but the fundamentals of technical SEO. However, the link between “top 10” and “citation” is loosening after query fan-out, so emphasis is shifting from single-keyword rankings toward coverage of the whole topic cluster.
2. Bot accessibility is the first gate of GEO and the largest difference from SEO. Googlebot renders JavaScript, but none of the AI crawlers do. A CSR site may be visible to Google and still be an empty page to an LLM. This means React-based sites built with vibe coding may be structurally disadvantaged in GEO, which makes it a direct topic for no-code and vibe-coding education.
3. Measure with the MSC funnel. Breaking the vague question “Does AI recommend us?” into three stages (mention, source, citation) plus a matrix by engine pinpoints where to improve. The framework can also be used for manual checks without any GEO tool.
4. Search engines already publish the clues for content. PAA questions, the domains cited in AI Overviews, and related keywords form the content backlog. Deciding what to write is a research problem more than a creative one.

Sources

• YouTube interview (primary source and backbone of this research): Hawaii podcast, “GEO optimization to get GPT and Claude to recommend your brand” (CEO Lee Jae-cheol of HaloX), https://youtu.be/6S_3OfKM-No. Full transcript: interview-transcript.txt (accessed 2026-08-20, subtitles extracted with yt-dlp)
• GEO paper (primary): Aggarwal et al., “GEO: Generative Engine Optimization,” arXiv 2311.09735 / KDD 2024, https://arxiv.org/pdf/2311.09735 · https://dl.acm.org/doi/10.1145/3637528.3671900 (accessed 2026-08-20)
• AI crawlers and JavaScript (primary research): Vercel × MERJ, “The rise of the AI crawler,” https://vercel.com/blog/the-rise-of-the-ai-crawler (accessed 2026-08-20)
• AI Overview citations vs. top 10 (secondary, cross-checked across several studies):
  • Search Engine Journal, report on Ahrefs’ drop from 76% to 38%, https://www.searchenginejournal.com/google-ai-overview-citations-from-top-ranking-pages-drop-sharply/568637/ (accessed 2026-08-20)
  • BrightEdge, 54% from organic ranking, https://www.brightedge.com/resources/weekly-ai-search-insights/rank-overlap-after-16-months-of-aio (accessed 2026-08-20)
  • Originality.AI, top 10 at 52%, https://originality.ai/blog/google-ranking-ai-citations-study (accessed 2026-08-20)
• Caveat: the MSC framework, the claims about domains and sentences preferred by AI, and the commerce text-embedding workaround come from the interviewee’s field observations (HaloX) and have not been independently verified. When citing them externally, label them as “practitioner observations.”

Bottom line

GEO does not start with new tricks. It starts with the technical SEO basics that make a site readable by bots, and the evidence says the most important of these for LLMs is rendering. AI crawlers do not execute JavaScript, so content that only appears after client-side rendering is invisible to them, while Google can still see it. Ranking in the top 10 no longer guarantees an AI citation, so topic coverage and measurement through the mention, source, and citation stages matter more than any single keyword position.