Ethan Smith on Answer Engine Optimisation, LLM Search, and the Citation Playbook

Ethan Smith on Answer Engine Optimisation, LLM Search, and the Citation Playbook

transcriptlennyethan-smithseoaeollmsearchgrowthcitationsgraphite

Ethan Smith on Answer Engine Optimisation, LLM Search, and the Citation Playbook

Ethan Smith — Lenny’s Podcast · ~2025 · Source

Ethan Smith, CEO of Graphite and 18-year SEO practitioner, explains how to get your product to show up in LLM-generated answers (ChatGPT, Perplexity, Gemini, Claude). AEO (Answer Engine Optimisation) is the second-biggest shift in search since Google’s spam-prevention algorithms of the mid-2000s. Illustrated with Webflow client data: 6× conversion rate from LLM vs. Google traffic; 8% of signups now from LLMs. The episode also covers a controlled study on AI-generated content (it doesn’t rank), a model-collapse experiment, and a counterintuitive finding about help centre optimisation.

Key ideas

  • AEO vs. SEO: same foundation, different head and tail. AEO operates on the RAG layer (LLM + web search → summary). Everything that works in SEO also works in AEO. The head is different: citation count determines ranking, not citation rank — so a new company mentioned in 10 Reddit threads beats an incumbent with one #1-ranked page. The tail is different: average query is ~25 words (vs. ~6 in Google), many queries have never been asked before, and early-stage companies can win immediately without domain authority.

  • Seven-step AEO playbook. (1) Identify target questions (from paid search keywords → turn into questions); (2) set up answer tracking (share of voice, not rank); (3) audit which citations LLMs currently use; (4) create on-site content answering all follow-up questions; (5) build off-site citations by group — affiliates, YouTube/Vimeo, Reddit (real identity, useful answer, say where you work); (6) run controlled experiments with test vs. control question groups; (7) staff with SEO team for on-site + a community/marketing generalist for off-site.

  • AI-generated content does not rank. Smith’s team ran a controlled study: validated an AI detector (pre-ChatGPT false-positive baseline ~8%), sampled thousands of Google and ChatGPT citation URLs, found only 10–12% is AI-generated. Correlation analysis confirms AI content is not ranking. AI-assisted content does. The model-collapse argument: feeding LLM derivatives back into RAG converges wisdom of the crowd to a single opinion, destroying source diversity.

  • Help centre optimisation is an underutilised AEO asset. LLM tail queries often ask use-case and integration questions that only product help docs answer. Actions: move help centre to a subdirectory (not subdomain), add internal cross-links, and fill the tail by mining sales calls, customer support tickets, and Reddit threads for questions not yet answered anywhere.

  • Misinformation in AEO is pervasive. Google is not dying — Google VP of Search confirmed publisher traffic is flat or slightly up. New LLM surfaces add to the pie, they do not take from Google’s slice (same pattern as TikTok, Instagram, YouTube search before them). Most AEO best practices published online have not been verified by controlled experiment; reproduce before adopting.