Introduction

Search has changed. LLMs no longer rank pages the old way but create answers, which rewrites how visibility works.

What's the problem it solves?

People are arguing if AI search is just old SEO with a new name. This article shows the real issue: the math behind search has changed so much that old SEO rules no longer map cleanly to how LLMs produce answers.

Quick Summary

For 25 years, search worked the same way. You typed a query, and the engine ranked pages. SEO lived inside that system: match intent, rank high, get clicks. Even when Google added things like knowledge panels or instant answers, the math behind ranking did not change. It was still one goal: pick the best set of documents and place them in order.

LLM search is different. It does not return a ranked list. It builds an answer using pieces of text pulled from many sources. That means the system now has two steps: retrieve chunks and generate an answer. Because of this, the math behind what gets chosen has shifted. Content must first be retrieved, then actually shape the final answer. This creates a brand new visibility problem.

The article shows two real SEO strategies. Before LLMs, big pillar pages tended to rank best. With LLMs, short pages with one clear focus tend to get pulled into answers more often. This is because long pages get penalized in retrieval, mix topics, and lose clean matches in both sparse and dense scoring. The result is simple: what wins in old SEO does not always win in AI answers.

This is the biggest change in search because the output type, the scoring goal, and even the behavior of the system (now probabilistic, not deterministic) have all shifted. No past Google update changed the math this much.

Key Takeaways

• Search is no longer only about ranking pages. LLMs build answers from text chunks.
• Old SEO and LLM visibility optimize different math goals. Doing well in one does not ensure success in the other.
• Big pillar pages rank well in classic SEO but are less likely to be pulled into LLM answers.
• Short, focused pages with one clear topic tend to be retrieved more often by LLMs.
• Retrieval algorithms punish long mixed-topic pages through length limits, embedding drift, and chunk scoring.
• LLM answers are not deterministic, which breaks the old idea of stable rankings.
• This shift is the biggest change in search since search began.

What to do

• Create more short, focused pages that answer one question clearly.
• Put the key answer at the top of the page so it sits in the first chunk.
• Avoid mixing too many topics inside one long guide if you want LLM visibility.
• Keep pillar pages, but pair them with scoped pages for each core question.
• Write in clean, simple language so embeddings match the query more directly.
• Track how your pages show up in AI answers, not only in classic rankings.
• Treat LLM visibility as its own goal, not a side effect of traditional SEO.

‍