GEO: Optimizing for LLM retrieval, not just Google
How we restructured content for ChatGPT, Perplexity, and Claude web search. Semantic density beats keyword density.
Search is splitting in two. Traditional Google SEO still matters, but a growing percentage of our clients' qualified traffic now comes from LLM-based search: ChatGPT with browsing, Perplexity, Claude web search, and Gemini.
These systems do not rank pages with PageRank. They retrieve and synthesize. Which means the optimization playbook is completely different.
How LLMs retrieve content
Large language models with web access typically:
- Rewrite the user query into search terms
- Fetch top search results
- Chunk and embed the content
- Retrieve relevant chunks via vector similarity
- Synthesize an answer with citations
This means your content must be retrievable and synthesizable.
Semantic density over keyword density
Traditional SEO teaches keyword placement: H1, first paragraph, image alt text, meta description. LLMs do not read meta descriptions. They read the actual text, chunk it, and compare embeddings.
What works:
- Clear topical boundaries: Each page should have a distinct, well-defined subject. LLMs struggle with pages that cover five unrelated topics.
- Entity-rich copy: Use precise nouns. "PostgreSQL 16" is more retrievable than "our database."
- Structured arguments: Use headings, lists, and tables. These create natural chunk boundaries.
- FAQ sections: Direct question-answer pairs match LLM query patterns exactly.
- Citations and sources: LLMs prefer content that references authoritative sources.
What we changed
For three client sites, we restructured content with LLM retrieval in mind:
- Split broad pages into focused, single-topic pages
- Added explicit definitions for technical terms
- Restructured H2s as questions users actually ask
- Included comparison tables for products and services
- Added author bios and expertise signals for E-E-A-T
Results
After 90 days:
- ChatGPT citations increased 340%
- Perplexity referral traffic up 280%
- Brand mention rate in AI answers increased from 12% to 41%
- Traditional Google rankings improved as well (structured content helps both)
The future
We believe GEO will become a standard discipline alongside SEO. The fundamentals overlap: clear structure, fast loading, authoritative content. But the execution differs.
Start by auditing your content for semantic clarity. If an LLM cannot chunk and retrieve your key points, you are invisible to the fastest-growing search channel.