CONCEPT

GEO / AEO (Getting Recommended by AI)

Properties1
tagsgeo, aeo, ai-search, llm, generative-engine-optimization, answer-engine-optimization, ai-recommendation

5 min read

GEO / AEO (Getting Recommended by AI)

Optimizing so that AI assistants (ChatGPT, Claude, Gemini, Grok, Perplexity) recommend your business — Generative Engine Optimization / Answer Engine Optimization.

Definition

GEO/AEO is the practice of influencing what AI models say when a user asks for a recommendation (“best plumber in Gary”). Per Caleb Ulku, models have no internal business database: they either search the web in real time and synthesize from what they find (Reddit, forums, Medium, review sites) or, for well-known entities, answer from training memory. Crucially, models have no sense of content age — old complaints weigh as much as fresh reviews. The goal is to move a brand from “looked up and guessed at” to “recommended from memory,” and to control what real-time search surfaces.

The proposed mechanism is consensus: an AI trusts information that recurs across multiple trusted sources, formats, and perspectives, and discounts single-source echo chambers. The tactic is a diverse content footprint (see 250 Authority Protocol) plus the same entity alignment that wins Google, since AI models read similar signals. Consensus level is emerging: the direction is credible and increasingly important, but specific tactics (and the training-data-influence claims) are unproven and evolving.

Independent corroboration + data (Dooley). James Dooley adds practitioner data from an independent source: direct LLM-referred leads are still tiny (~1.3–1.4%) but branded search (driven by AI Overviews/citations) is ~15–20% and converts far better — so the near-term GEO payoff is branded-search lift, not direct AI leads. He recommends LLM seeding via Semantic Triples (Entity–Attribute–Value), listicles, and third-party corroboration (reviews/awards), and predicts Gemini/Claude overtake ChatGPT on Google’s data moat. Nico (AI Ranking) ties review wording (service + location) to “ask maps mode.”

Independent data (Ahrefs, 2026). AI SEO Course for Beginners: Complete AEO Tutorial (Ahrefs) (Sam Oh) is the first non-Caleb, measurement-heavy AEO source. AEO builds on SEO; it does not replace it. AI Overviews cut #1 CTR ~58% (Dec 2025). Volume is tiny (industry ~0.25% of traffic; Google still ~210× AI platforms) but conversion is fat (Ahrefs: 0.5% traffic → 12.1% sign-ups). Strongest lever in a 75k-brand study: branded web mentions (0.664 corr. with AI Overview visibility) — stronger than backlinks/DR. Freshness is first-class (ChatGPT: 76% of top cited pages updated in 30 days) — direct tension with Caleb’s “models have no sense of content age.” Word count correlation 0.04. YouTube is a sleeper channel. ChatGPT’s crawler does not render JS. Their fake-brand plant: Gemini/Perplexity repeated fiction 37–39%; ChatGPT stayed <7%.

Caleb course addendum. How Google and AI Actually Work claims ChatGPT relies on Bing’s index (Microsoft/OpenAI), so a Bing Places profile matters more than optimizing for “other AI bots.” Independent coverage of that claim is still thin in-wiki (Search Engine Land 2025 nuance: ChatGPT uses Bing web results, not Bing Places profile details directly — ingest that primary when the source page lands).

Cloudflare economics. Making AI search smarter cites Pew 2025: when an AI summary appears, users clicked a traditional result 8% of the time (about half the no-summary rate) and a link inside the summary only 1%. Pay-per-use experiments are the publisher-side of AEO.

Reality-check on the training-data claim. The 250 Authority Protocol’s “get into the model’s training data” mechanism is not supported by the primary Anthropic study it cites (see Data Poisoning & LLM Backdoors). Ahrefs’ own framing is mixed: they still talk about training-data cadence (~6 months) and RAG/retrieval. GEO that works via real-time retrieval, mentions, and consensus is the defensible lane; GEO that claims to implant brand memory in training is unproven. Consensus stays emerging, but it is no longer single-voice.

Key Properties

  • Two AI modes: real-time web search vs. answering from training memory — GEO targets both.
  • Models over-index on Reddit/forums/reviews and have no time-awareness (stale content persists).
  • Wins via consensus + diversity of content, not single high-authority pages.
  • Convergent with local SEO: entity alignment (GBP–Website Alignment (Consistency Signals)) helps both; “sell AI presence, not just SEO.”
  • Emerging + partly speculative — see the honest caveats logged in TRAIN Any AI To Recommend You (This Study Proves How).

Examples from Sources

ExampleSource
”Model Training Data Risk Auditor” prompt surfaces what Reddit/reviews currently say about a brandTRAIN Any AI To Recommend You (This Study Proves How)
Fixing entity alignment for Google simultaneously wins ChatGPT/Perplexity/ClaudeThis AI Tool Will Make You RANK FAST (Full Walkthrough)

In the Sources

SourceContext
TRAIN Any AI To Recommend You (This Study Proves How)The core GEO source: real-time vs. memory, consensus mechanism, the 250 Authority Protocol.
This AI Tool Will Make You RANK FAST (Full Walkthrough)AI models evaluate the same entity signals as Google.
2026 SEO Secrets That Ranked 1,000+ Sites (James Dooley, interviewed by Sterling Sky)Independent data: branded-search lift vs. tiny direct-LLM leads; LLM seeding; Gemini/Claude trends.
Local SEO was hard until I built systems like this (99+ bookings/mo)Review wording (service + location) for “ask maps mode.”

Concepts: 250 Authority Protocol, Data Poisoning & LLM Backdoors, Semantic Triples (Entity–Attribute–Value), Social-Media Parasite SEO, Brand & User Signals as Ranking Drivers, Entity-Based SEO, Local SEO Entities: ChatGPT, Claude, Perplexity, Gemini, Anthropic, Reddit, Google, James Dooley