Concept

Answer Engine Optimisation

conceptseoaeollmsearchgrowthmarketing

Answer Engine Optimisation

Also written AEO (Answer Engine Optimization in American English). The discipline of making a product, brand, or content appear in LLM-generated answers from systems such as ChatGPT, Perplexity, Gemini, and Claude. Coined by practitioners to distinguish from SEO; Ethan Smith argues AEO and GEO (Generative Engine Optimisation) are synonyms, preferring AEO because “answer” is narrower than “generative.”

How it works

LLM answer engines operate primarily via RAG (retrieval-augmented generation): the model runs a web search, retrieves a set of citations, then summarises them. AEO targets this layer — not the model’s training weights — making it tractable and fast (unlike training data influence, which lags by months or years).

Critical difference from SEO: In Google, rank 1 = win. In LLMs, the product mentioned most often across citations is presented first, regardless of any individual citation’s rank. The LLM synthesises across sources.

Head vs. tail

SEOAEO
Head queriesRank determines outcome; domain authority is prerequisiteCitation count determines outcome; no domain authority required
Query length~6 words average~25 words average (Perplexity data)
Tail queriesExisting tail well-mapped via Google Keyword PlannerLarger tail; many queries never asked before; early-stage companies can win immediately

Citation groups

The off-site layer of AEO — the sources LLMs retrieve from:

  • Own site: traditional SEO pages + help centre docs (subdirectory, not subdomain; cross-linked; filled with tail questions)
  • Video: YouTube, Vimeo — B2B topic verticals have low competition in video; early opportunity
  • UGC: Reddit, Quora — highest trust signal for LLMs; real accounts, real identities, genuinely useful answers; spam is detected and removed
  • Tier-one affiliates: Dotdash Meredith (Allrecipes, Good Housekeeping, Investopedia), TechRadar for B2B — most-cited affiliate class in LLMs
  • Tier-two affiliates and blogs: paid placement or earned coverage

Where mainstream views differ

The dominant narrative (mid-2024 onwards) frames AEO as a break with SEO — a new discipline requiring new expertise and tools. Smith’s counter-thesis: everything that works in SEO works in AEO; the differences are additive, not substitutive. Citation optimisation and the longer tail are the genuine additions; on-site content strategy is largely the same.

A second contested claim: “just write great content, the algorithms will find it” (Nick Turley, Head of ChatGPT, on the same podcast channel). Smith disagrees in principle while agreeing that spam strategies will fail — the practical difference is that authoritative content alone is necessary but not sufficient; active citation-building accelerates ranking and is measurable.

AI-generated content

Smith’s study (thousands of URLs, validated AI detector, ~8% false-positive baseline): ~10–12% of content in Google Search and ChatGPT citations is AI-generated. 90% is human-written. AI-assisted content ranks; 100% automated AI content does not. Model-collapse theory explains why: if LLM derivatives are fed back into RAG citations and re-summarised, the system converges on a single opinion, destroying the diversity that makes the wisdom-of-the-crowd effect function.

Measurement

Share of voice: percentage of LLM responses that mention your brand, across a set of target questions. The correct metric; replaces rank. Requires tracking multiple question variants and multiple surfaces (ChatGPT, Perplexity, Gemini, Meta AI) because LLM answers are probabilistic.

Experiment design: split target questions into test and control groups; intervene on test only; compare share of voice before and after; reproduce before treating the result as settled.