The AEO and LLM SEO space is thick with jargon, and much of it is used loosely or sold as magic. This glossary defines 51 terms across five groups: the disciplines, the mechanics, the surfaces, the technical levers, and the measurement metrics. Each definition is written to be honest about what the evidence actually supports, not what vendors claim.
LLM SEO
LLM SEO is optimizing your content and site so language model search engines can retrieve, cite, and recommend you. In practice it is conventional SEO plus retrievability, not a separate discipline. Google states that optimizing for generative AI search is still SEO.
AEO (Answer Engine Optimization)
AEO is optimizing to be the sourced answer inside answer engines like ChatGPT and Perplexity, rather than a blue link. It is functionally a synonym of LLM SEO and GEO. See how AEO, SEO, and GEO fit together.
GEO (Generative Engine Optimization)
GEO is the academic name for tailoring content to appear in generative answers, coined in a KDD 2024 paper. Treat its specific tactics with caution, because they failed independent replication in C-SEO Bench, which rated traditional SEO significantly more effective. It is covered in depth in what GEO is.
LLMO (Large Language Model Optimization)
LLMO is yet another label for the same practice, emphasizing optimization for the model layer. It carries no distinct, evidenced method set of its own.
AI Search Optimization
AI search optimization is the umbrella term for optimizing across all AI search surfaces, from AI Overviews to ChatGPT to Perplexity. The evidenced levers under it are retrievability, notability, and conventional SEO. It is broken down in the AI search optimization playbook.
Generative Search
Generative search is a search experience that returns a synthesized natural language answer, often with citations, instead of a ranked list of links. It is the category that AEO and GEO target.
Answer Engine
An answer engine responds to a query with a direct synthesized answer and cited sources, rather than a list of results. Perplexity and ChatGPT Search are examples. The discipline AEO is named after it.
Query Fan-out
Query fan-out is when an AI system takes one query and issues a set of concurrent, related sub queries, then synthesizes results across all of them. This is the biggest real shift in the field. You now compete for many sub questions, not one head term.
Retrieval-Augmented Generation (RAG)
RAG is an architecture that retrieves relevant documents from an external source and injects them into the model's context before it generates. It grounds the answer in fetched text rather than only the model's memory. Most AI search runs on RAG.
Grounding
Grounding ties a model's output to verifiable external sources or live data, so claims can be attributed and are less likely to be fabricated. Retrieval is how grounding is usually achieved.
Retrievability
Retrievability is whether a page can be fetched into an AI engine's context at all: crawlable, indexable, server rendered, not blocked. It is the evidenced precondition for citation. No retrieval means no mention, whatever else you do.
Site Notability
Site notability is how well known and widely referenced a domain is, and it is the strongest observed correlate of getting cited. In a study across 55,936 queries and six engines, global site popularity was the single most influential feature in the model. The honest takeaway is to be a genuinely notable brand, which is PR and product, not markup.
Citations vs Mentions
A citation is a linked source attribution in an AI answer; a mention is your brand named in the prose without a link. They are different outcomes with different levers. Only about 6 to 27 percent of the most mentioned brands are also top cited sources.
Extractability
Extractability is structuring content so a specific passage, statistic, or definition can be lifted cleanly into an answer once the page is retrieved. It helps you get quoted once found, not found in the first place. Do it because it is good writing, not as a visibility strategy.
Hallucination
A hallucination is fluent, confident output that is factually wrong or unsupported by any source. Grounding and retrieval reduce it but do not eliminate it.
Zero-click Search
A zero-click search ends without a click to any external site because the answer is shown on the results surface itself. Pew found clicks fall from 15 percent to 8 percent when an AI summary appears, and only 1 percent click a link inside it. The goal shifts from traffic to influence within the answer.
Prompt
A prompt is the natural language question or instruction a user types into an AI engine. In AEO, real buyer prompts, not brand name lookups, are the demand you measure and target.
Token
A token is the unit of text a language model processes, roughly a word or word piece. Model limits and pricing are counted in tokens.
Embedding
An embedding is a numeric vector that represents the meaning of a piece of text, so similar meanings sit close together in vector space. Retrieval systems use embeddings to find relevant passages.
Vector Database
A vector database stores embeddings and returns the closest matches to a query vector at speed. It is the retrieval backbone behind many RAG systems.
Semantic Search
Semantic search matches on meaning rather than exact keywords, using embeddings to find conceptually related content. It is how AI engines locate relevant passages even when the wording differs.
Large Language Model (LLM)
An LLM is a model trained on vast text to predict and generate language, and to reason over it. GPT, Gemini, and Claude are examples that power today's AI search.
Context Window
The context window is the amount of text a model can consider at once, measured in tokens. Retrieved documents must fit inside it to influence the answer.
Share of Voice (AI)
AI share of voice is the percentage of AI answers, across a defined prompt set, in which your brand appears, measured against competitors. It is meaningless from a single run, so measure it across repeated runs.
Corroboration
Corroboration is when multiple independent, trusted sources describe your brand or claim the same way. Generative engines lean on consensus, so the description many sources agree on is the one the model repeats.
AI Overviews
AI Overviews are Google's AI summary at the top of the results page, powered by Gemini and built via query fan-out. They are served by Googlebot, not a separate AI crawler, and grounded in the core Search index.
AI Mode
AI Mode is Google's dedicated conversational search experience, with follow ups and heavy query fan-out, powered by a custom Gemini model. It is rooted in the same core Search ranking systems.
ChatGPT Search
ChatGPT Search is OpenAI's web search inside ChatGPT that retrieves live results and returns cited answers. Its indexing crawler is OAI-SearchBot, which is distinct from the training only GPTBot.
Perplexity
Perplexity is an answer engine that responds to queries with synthesized, citation backed answers using its own retrieval and index. Its crawler is PerplexityBot, and it aligns with search rankings more than other engines do.
Microsoft Copilot
Copilot is Microsoft's generative assistant, grounded in the Bing index. Bing Webmaster Tools is currently one of the only free windows into real AI grounding and fan-out queries.
Google Gemini
Gemini is Google's standalone assistant app, built on the Gemini model family with optional Google Search grounding. It is a separate product from the AI features inside Search.
Grok
Grok is xAI's assistant, native to X, whose edge is real time grounding on the X firehose plus live web search. It is tuned for current events and a less filtered tone.
AI Crawlers / Bots
AI crawlers are the user agents AI platforms use to fetch web content. The critical distinction is that retrieval and search bots feed citations while training bots feed model weights. Decouple them, and verify a bot by its published IP range, not its user agent string.
GPTBot
GPTBot is OpenAI's training crawler, which gathers content to improve future models. Blocking it affects training, not whether ChatGPT Search can cite you.
OAI-SearchBot
OAI-SearchBot is OpenAI's search crawler that indexes pages so ChatGPT Search can retrieve and cite them. This is the bot that matters for ChatGPT citations, so allow it.
Google-Extended
Google-Extended is a control token for Gemini training and grounding, not a crawler. Blocking it does not remove you from AI Overviews, which ride Googlebot, so it is a common and expensive null action.
PerplexityBot
PerplexityBot is Perplexity's crawler that indexes pages for its answer engine. Allowing it is a prerequisite for being cited there.
ClaudeBot
ClaudeBot is Anthropic's training crawler. Anthropic runs a separate search fetcher for Claude's web search, so treat training and search access as different decisions.
robots.txt
robots.txt is the file that tells compliant crawlers which paths they may fetch. It is a crawling directive, not an access control or a deindex tool, and user triggered agents may ignore it. See robots.txt for AI bots.
llms.txt
llms.txt is a proposed markdown file that lists a site's key content to help LLMs. No major AI engine uses it. Google says it neither helps nor harms, and one study found about 97 percent of these files are never even fetched.
Schema / Structured Data
Schema is machine readable markup that describes a page's entities, earning rich results and aiding entity disambiguation. It is not a proven AI citation lever. A controlled test found near zero effect on citations and a small decline on AI Overviews, and FAQ rich results were retired on 2026-05-07.
JSON-LD
JSON-LD is the recommended format for adding schema.org structured data as a script block. Engines tend to read it as plain text rather than parsing it as schema, so it is not a shortcut to citation.
Core Web Vitals
Core Web Vitals are Google's field metrics for user experience: LCP, INP, and CLS. They carry only a small ranking effect, so justify them on user experience and conversion, not on AI citations.
Indexability
Indexability is whether a page can be crawled and included in an index, governed by robots.txt, noindex, canonicals, and render dependency. It is more load bearing for AI search than classic SEO, because most AI crawlers do not run JavaScript.
Passage Ranking
Passage ranking is Google ranking a specific passage within a longer page for a query. Note the myth: there is no passage indexing. Google has said plainly it does not index passages, so there is no need to chop content into tiny chunks.
Entity
An entity is a distinct thing, such as a brand, person, or product, that a machine can identify by its attributes and relations. Strong entity signals help AI recognize and describe you correctly.
Knowledge Graph
The Knowledge Graph is Google's database of entities and their relationships. A Knowledge Panel appearing for your brand is visible confirmation that Google treats you as a recognized entity.
E-E-A-T
E-E-A-T stands for Experience, Expertise, Authoritativeness, and Trust, a concept from Google's rater guidelines. It is not a ranking factor. Google states this verbatim, so treat any product selling E-E-A-T optimization with skepticism.
AI Referral Traffic
AI referral traffic is website sessions whose referrer is an AI surface, such as chatgpt.com or perplexity.ai. It is real but tiny today, typically a fraction of one percent of sessions, so do not over prioritize it.
Prompt Tracking
Prompt tracking is monitoring a fixed set of representative prompts across engines over time, to see whether and how a brand is mentioned or cited. It is the standard AEO measurement method, and it is only meaningful with repeat run averaging.
Measurement Variance
Measurement variance is the run to run instability of AI answers: identical queries return different results. Commercial engines return an overlap of roughly 0.34 to 0.42 across repeats, so any single shot visibility score with no confidence interval is reporting noise as signal.
Frequently Asked Questions
What is LLM SEO?
LLM SEO is the practice of optimizing content and site infrastructure so that language model powered search engines can retrieve, cite, and recommend you. In practice it is mostly conventional SEO plus retrievability, not a separate discipline. Google states that optimizing for generative AI search is still SEO, drawn from the same index and quality signals.
Is LLM SEO the same as AEO and GEO?
Largely yes. LLM SEO, AEO, GEO, and LLMO are competing labels for the same discipline: getting surfaced and cited in AI answers. The names differ, the practice does not. No independent evidence shows any of them beats good SEO, and the academic GEO tactics failed independent replication in C-SEO Bench at NeurIPS 2025.
What is LLMO?
LLMO, or Large Language Model Optimization, is another synonym for optimizing content so AI models and the search products built on them mention and cite you. It describes the same goal as AEO, GEO, and LLM SEO. The label is recent and marketing coined, and it carries no distinct, evidenced method of its own.
Does schema markup help you get cited by AI?
Not as a citation lever. Structured data earns rich results and helps disambiguate entities, but a controlled test found near zero effect on ChatGPT and AI Mode citations, and a small decline on AI Overviews. Google says structured data is not required for generative AI search. Keep it for rich results, not AI visibility.
Does llms.txt improve AI visibility?
No. No major AI engine consumes llms.txt today, and Google says it neither helps nor harms. A study of many thousands of domains found that about 97 percent of llms.txt files were never even requested, mostly by SEO audit tools checking whether one exists. It is not a working visibility lever.
If I rank number one on Google, will ChatGPT cite me?
Usually not. Only about 8 percent of ChatGPT citations come from Google's top ten results. AI engines assemble citations through query fan-out across many sub questions, and they pull from a broader, different source set, so classic rank position is a weak predictor of being cited in an AI answer.
How do you measure AI visibility reliably?
Track a fixed set of prompts across the engines over time, but average 5 to 10 runs per prompt. Identical queries return different answers on repeat, with overlap around 0.34 to 0.42, so any single run score without a confidence interval is reporting noise rather than signal. Separate mentions from linked citations too.
Where Unveilr fits
A glossary tells you what the terms mean. The harder job is knowing which of them actually move your visibility, and then moving it. Unveilr runs that loop: scan how AI engines answer your priority prompts, detect where you are missing or losing ground, update the content and signals that close the gap, then re-scan to confirm the lift is real rather than noise.
That last step matters, because the field is full of tactics that sound rigorous and do nothing. In one internal case study, a D2C brand moved from the ninth most-cited domain to the single most-cited source in AI answers, with its ChatGPT visibility rising from 3.3 percent to 44.7 percent.