Answer Engine Optimization for LLM Retrieval | Sergey Lucktinov

With Sergey Lucktinov, Digital strategist and author of Semantic SEO, SRO and AI, with about 16 years in SEO and a focus on GEO and AEO · Streamed live · 54 min

What this episode covers

Sergey Lucktinov explains how LLM retrieval works, describing fan-out queries, metadata filtering and passage reranking, and how this shapes answer engine optimization. He argues that sites should be built as semantic content networks, with fast load times, answer-first passages, self-contained focused pages and one query intent per page. The main conclusion is that clear, on-topic, well-structured content matching customer questions is what gets retrieved into AI answers.

Key takeaways

  • ✦Keep time to first byte under 300 milliseconds, since slower sites may not be checked by LLM retrieval at all.
  • ✦Answer the target question in the first 50 to 70 words of each relevant section.
  • ✦Keep each passage self-contained and on one topic, avoiding long paragraphs that mix concepts.
  • ✦Target one query intent per page and create separate pages for definitions, comparisons and reviews.
  • ✦Structure the site as a semantic content network linking macro, seed and node pages, and avoid popups that disrupt the content.

Key moments

  1. 21:27What SRO Means for Retrieval
    “the main concept in uh LLM retrieval right now is a cost of retrieval”
  2. 28:04Speed and Answer-First Structure
    “make sure that your website um um time to first bite one of the uh most important um uh KPIs”
  3. 31:53Macro and Micro Semantic Structure
    “There's there are two different things uh two major concepts in semantic SEO. This is macro semantics and microsemantics.”
  4. 34:01Building a Semantic Content Network
    “You never link from uh macra to uh node or from node to macra because this breaks a semantic uh semantic relationship.”
  5. 45:32Popups and Trust Signals
    “the website should not switch uh elements uh like the website should be uh uh static”
  6. 48:24One Query Intent Per Page
    “the golden rule uh LLM rule is uh one query type per page”

Quotes are from the episode transcript, lightly cleaned of auto-caption errors.

Questions answered in this episode

How fast should a website be for LLM retrieval?
Sergey says time to first byte should be less than 300 milliseconds, because if it is longer LLM will not check the website at that stage.
How many words should the answer come first in a section?
He advises answering the target question in the first 50 to 70 words of each relevant section.
Should one page target multiple query types?
No, he says the golden rule is one query type per page, and other intents should get their own article.

More episodes on AI Search & LLMs