SOURCE

AI SEO Course for Beginners: Complete AEO Tutorial (Ahrefs)

Properties1
tagsseo, geo-aeo, aeo, brand-mentions, query-fan-out, ahrefs, chatgpt, perplexity, youtube
AUTHOR
Sam Oh (Ahrefs)
PUBLISHED
2026-06-17
ORIGINAL
view original

7 min read

AI SEO Course for Beginners: Complete AEO Tutorial

Author: Sam Oh / Ahrefs (NEW) | Published: 2026-06-17 | Source: transcript (86:56, 73K views)

Summary

Ahrefs’ four-module AEO course (Sam Oh) is the wiki’s first independent, data-backed GEO / AEO (Getting Recommended by AI) method — SaaS/brand AEO, not local. AEO = making content visible and citable to AI systems that answer directly (Google AI Overviews/AI Mode, ChatGPT, Perplexity, Gemini, Copilot). He treats GEO/LLMO as synonyms and insists AEO builds on SEO, it does not replace it.

The urgency numbers: as of Dec 2025 an AI Overview cuts CTR for the #1 organic page by 58%; ChatGPT has ~900M weekly users (~12% of Google volume); AI referral traffic grew 9.7× in a year. Volume is still tiny — Ahrefs’ own AI search was 0.5% of traffic in June 2025 but 12.1% of sign-ups (23× organic conversion); later they cite ~0.25% of average site traffic and Google still sending ~210× the top AI platforms combined. Vercel (10% conversion) and Tally (+$1M ARR) are the other named proof points. Direct LLM clicks undercount because many platforms strip referrers; self-reported “how did you hear about us” is the revenue metric (~3% of Ahrefs conversions).

Mechanics: models pull from training data (static, ~6-month cadence) and RAG / real-time retrieval. One prompt fans out into many synthetic sub-queries (Seer/Nectiv: avg 9–11, some 28; ChatGPT Deep Research ran 420 searches for one shopping query) — 95%+ have zero search volume, so do not treat fan-outs as a keyword list. Citations are probabilistic (temperature), not ranked. Consensus, freshness (~25% fresher than classic SERPs; ChatGPT: 89.7% of top cited pages updated in 2025, 76% within 30 days), and authority all matter. An earlier Ahrefs figure that 76% of AIO citations came from Google’s top 10 is walked back to ~38%; 14% of AIO-cited pages are not even in the top 100. Platform overlap is thin: only 7 of the top 50 cited domains appear on AIO, ChatGPT, and Perplexity (14%). YouTube is ~5.6% of AIO citations and the #1 AI Mode domain; YouTube mentions correlate 0.737 with ChatGPT visibility (GPT-4 trained on 1M+ hours of transcripts). ChatGPT prefers high-DR publishers (median DR 90; Reddit/Wikipedia/Forbes); Perplexity is closest to classic Google (28.6% of citations from the top 10 vs ChatGPT 8–10%).

The strongest lever in their 75k-brand study is branded web mentions (0.664 correlation with AIO visibility) — stronger than backlinks, DR, or referring domains; mentions on highly-linked pages rise to 0.7. Only ~28% of mentions include a link (AIO 10.7%, ChatGPT 26.9%, Perplexity 51.6%). Execution: BLUF + atomic H2s + entity-rich SVO writing; word count correlation 0.04 (53.4% of cited pages <1,000 words); 43.8% of cited pages are listicles; brand-name your original frameworks so models cannot flatten them; refresh “sleeper” pages that already have links. Mentions in three tiers: editorial listicles/reviews, UGC (Reddit/Quora — no spam), owned properties (YouTube/podcast/LinkedIn). Technical: 5.9% of 140M sites block GPTBot; Cloudflare’s AI-bot robots.txt is on by default; ChatGPT’s crawler does not render JS (SSR required); llms.txt is unsupported by major providers; Schema Markup (Structured Data) evidence for AEO is mixed. They planted a fake luxury brand across a blog, Reddit, and Medium: Gemini/Perplexity repeated the fiction in 37–39% of answers; ChatGPT stayed <7% and cited the official FAQ 84% of the time — specific official content beats vague truth.

Key Claims

  • AEO = SEO + mention/citation competition, not a replacement. AI Overviews cut #1 CTR ~58% (Dec 2025). (Ahrefs data)
  • Tiny volume, fat conversion: Ahrefs 0.5% traffic → 12.1% sign-ups (23×); industry avg ~0.25% of traffic; Google still ~210× AI platforms. (first-party + cited case studies)
  • Two influence paths: get baked into training via wide consistent mentions, and rank in the retrieval set (classic SEO). Query fan-out (9–11+ synthetic sub-queries) is why topic coverage beats one-keyword pages.
  • Branded mentions beat links/DR (0.664; 0.7 on highly-linked pages) across 75k brands. ~28% of mentions include a clickable citation.
  • Freshness is a first-class AEO signal (~25% fresher; ChatGPT 76% of top cited pages refreshed in 30 days). Direct tension with Caleb’s “models have no sense of content age.”
  • Word count does not predict citation (r=0.04). Listicles = 43.8% of cited pages. BLUF, atomic sections, entity-rich declarative prose.
  • YouTube is the sleeper AEO channel (most-cited AIO domain; 0.737 corr. with ChatGPT). Optimize search-hit videos (keyword in title + spoken audio), not viral hits.
  • Technical access is table stakes: don’t block GPTBot/ClaudeBot/Google-Extended; SSR for ChatGPT; speed can drop a page from RAG; schema unproven for AEO; redirect hallucinated URLs (AI 404s 2.87× Google).
  • Misinfo is easy to plant on Gemini/Perplexity (37–39%); fill gaps with specific official FAQs. (Ahrefs experiment)
  • Measure three ways: AI referral (undercounts), bot hits, self-reported attribution. Monthly Brand Radar / quarterly competitive audit.

Notable quotes

“In June 2025, AI search accounted for just 0.5% of our traffic, but it drove 12.1% of our sign-ups. That’s a 23 times higher conversion rate than organic search.”

Why citable: Independent first-party confirmation of the Dooley pattern (tiny direct LLM share, outsized downstream value).

“Branded web mentions had the strongest correlation with AI visibility, a 0.664 correlation with showing up in AI overviews. That’s stronger than backlinks, domain rating, referring domains, or any other traditional SEO metric.”

Why citable: The course’s load-bearing empirical claim, and the cleanest contrast with link-first SEO and with Caleb’s 250-document training story.

“When AI has to choose between vague truth and specific fiction, it tends to choose the specific fiction.”

Why citable: Operational takeaway from their fake-brand planting test; pairs with Data Poisoning & LLM Backdoors without claiming you can buy your way into model weights.

Connections

Entities mentioned: Ahrefs (NEW), Sam Oh (NEW), Google, ChatGPT, Claude, Reddit, Perplexity, Gemini, Copilot, YouTube, OpenAI, Anthropic, Cloudflare, Graphite / Ethan Smith, Patrick Stokes, Tim Soulo, Ryan Law, Glen Allsopp, Vercel, Tally, Wirecutter, Wikipedia, Forbes, Quora Concepts referenced: GEO / AEO (Getting Recommended by AI), 250 Authority Protocol, Entity-Based SEO, Brand & User Signals as Ranking Drivers, Schema Markup (Structured Data), Data Poisoning & LLM Backdoors, Local SEO (by contrast — this is not a local playbook)

Contradictions / Tensions

  • vs. Caleb Ulku / 250 Authority Protocol on mechanism: Ahrefs never says “publish ~250 docs to enter training memory.” The lever is off-site branded mentions + retrieval SEO + freshness. Four-bucket diversity rhymes with their editorial/UGC/owned tiers, but the number, the Anthropic citation, and the “implant recommendation” claim do not survive this source. Treat Ahrefs as the source that can move GEO / AEO (Getting Recommended by AI) toward emerging/established on mentions+SEO, and further weaken the 250-as-science packaging.
  • vs. Caleb on content age: Caleb (TRAIN Any AI To Recommend You (This Study Proves How)) says models have no sense of age. Ahrefs’ citation set is materially fresher (25%+; ChatGPT 76% updated in 30 days). Named, testable disagreement — retrieval-time AEO clearly prefers fresh pages even if weights do not.
  • vs. Caleb on length / production: Ahrefs finds almost no word-count effect; 53% of cited pages <1k words. Caleb’s Core 30 / 8-pass pipeline produces long local pages. Different SERP (local pack vs. AI citation) may explain it — do not average.
  • vs. Schema Markup (Structured Data) as an AEO must: Ahrefs = mixed/unproven. Keep schema for SEO; do not sell it as an AEO ranking factor off this course.
  • Aligns with 2026 SEO Secrets That Ranked 1,000+ Sites (James Dooley, interviewed by Sterling Sky): listicles, third-party corroboration, branded-search > direct AI clicks, semantic/entity writing. Ahrefs supplies the numbers Dooley lacked.
  • Aligns with Data Poisoning & LLM Backdoors more than with the 250 protocol: their fake-brand test shows small planted corpora can distort Gemini/Perplexity answers — a retrieval/consensus failure, not a training-backdoor.
  • Commercial bias: the workflows run through Brand Radar, Keywords Explorer, Site Explorer, Bot Analytics. Keep the correlations and platform splits; treat “open this Ahrefs report” as product UX, not evidence.

Notes

Best current independent AEO source in the wiki. Use it as the default citation for “mentions > links for AI Overviews,” query fan-out, platform non-overlap, freshness, YouTube-as-AEO, and the 23× conversion pattern. Do not fold it into 250 Authority Protocol; it is a competing (more empirical) method. Ahrefs flagged NEW.