AEO has more vocabulary than SEO had in 2010. New acronyms arrive every quarter, some old ones get quietly redefined, and vendor-specific terms multiply across ChatGPT, Perplexity and Claude as well as Gemini and Copilot. If you've sat in a category strategy meeting where someone used AEO, GEO and LLMO in one sentence and then threw in MCP and RAG, this sheet is for you.
It holds 65 terms in seven groups, each explained in a line or two and current as of July 2026. Bookmark it and pass it to product and marketing, so nobody has to re-Google acronyms mid-meeting.
AEO fundamentals
People burn whole meetings arguing AEO versus GEO versus LLMO. I'd use AEO as the umbrella and move on, since the three overlap heavily and picking one rarely changes the work. The term worth being strict about is retrieval, because an engine can only cite what it fetched first.
| Term | Meaning |
|---|---|
| AEO (Answer Engine Optimization) | Structuring content so AI answer systems select it as a cited source; the umbrella discipline across generative and non-generative answer surfaces |
| GEO (Generative Engine Optimization) | The AEO subset focused on earning citations inside synthesized AI answers from ChatGPT, Perplexity and Google AI Overviews |
| LLMO (Large Language Model Optimization) | Widest of the three overlapping acronyms: any strategy for visibility inside LLM-based systems, including agents and enterprise chatbots |
| AIO (AI Optimization) | Informal catch-all used interchangeably with AEO or GEO, sometimes narrowed to Google AI Overviews inclusion |
| AI search | Information retrieval where a language model synthesizes a direct answer instead of ranked links; the query is treated as a prompt |
| Answer engine | A system built to return one synthesized response to a natural-language query, e.g. Perplexity, ChatGPT Search, Google AI Mode |
| Citation | An AI system's explicit reference to a source document in a response, which may appear as a footnote, inline link or source card |
| Retrieval | The pipeline step that fetches relevant documents before the model answers; its quality shapes which sources can appear |
| RAG (Retrieval-Augmented Generation) | Connects a language model to an external knowledge source at inference time, grounding output in retrieved documents instead of trained parameters |
| LLM (Large Language Model) | A neural network trained on large-scale text to predict and generate language; GPT-4o, Claude and Gemini are current examples |
| Foundation model | A large pre-trained base that developers fine-tune for specific applications; all LLMs are foundation models, and the category also covers image, audio and video models |
| Prompt | The natural-language input a user gives an answer engine, doing the job a keyword query does in traditional search |
AI search engines and interfaces
| Term | Meaning |
|---|---|
| ChatGPT Search | OpenAI's real-time web search inside ChatGPT, using the OAI-SearchBot crawler at query time; for Plus, Team and Enterprise subscribers |
| Perplexity | Standalone answer engine showing synthesized answers from live web results with numbered source cards; one of the highest-volume AI referrers to publisher sites |
| Claude with Search | Anthropic's web-retrieval mode, reaching current sources through the Claude-SearchBot crawler before generating a grounded answer |
| Gemini | Google's family of multimodal foundation models behind Google AI Overviews, Google AI Mode and the standalone Gemini assistant |
| Copilot | Microsoft's AI assistant in Bing, Microsoft 365 and Windows, built on OpenAI models with Bing's index as the retrieval layer |
| Google AI Mode | A separate chat-based interface within Google Search for complex queries; AI Mode and AI Overviews cite the same URLs only 13.7% of the time |
| Google AI Overviews | AI answer panels above standard Google results, evolved from the Search Generative Experience (SGE) experiment; 2.5 billion monthly active users by May 2026 |
| Grounded response | An AI answer whose every factual claim traces to a specific retrieved source, measured as the share of statements supported by retrieved context |
| Agentic search | An AI agent autonomously planning a multi-step research process, calling tools and fetching sources until it has enough evidence |
| Zero-click search | A session that ends on the results page with no click through to a source website; AI answers accelerate it |
Most of these you already know by name. The number I'd remember is 13.7%: Google runs two AI surfaces that agree on their sources only that often, so winning a citation in one tells you very little about the other.
200+ SaaS teams already track their AI citations.
They know exactly when ChatGPT mentions their brand, and when it stops. Do you?
Crawlers and indexing
Skim this group and you can do real damage. OpenAI and Anthropic keep training and retrieval apart (GPTBot versus OAI-SearchBot, ClaudeBot versus Claude-SearchBot), so you can refuse training without disappearing from answers, while Perplexity uses one crawler for both jobs and blocking PerplexityBot takes you out of its citations entirely. Syntax for each bot is in the full directive reference.
| Term | Meaning |
|---|---|
| GPTBot | OpenAI's training-data crawler, separate from OAI-SearchBot, which handles ChatGPT Search retrieval; blocking GPTBot affects training only |
| ClaudeBot | Anthropic's training-data crawler; Claude-SearchBot handles live Claude with Search retrieval under a separate user-agent string |
| PerplexityBot | Perplexity's crawler for both indexing and real-time fetching; blocking it removes a site from Perplexity's citation pool entirely |
| Google-Extended | A robots.txt opt-out token excluding content from Gemini training without affecting Google Search; no dedicated crawler behind it |
| Applebot-Extended | Apple's robots.txt token for excluding content from Apple Intelligence training while Applebot keeps indexing the site as usual |
| robots.txt | The plain-text file at a domain root that communicates crawl permissions to web robots via the Robots Exclusion Protocol |
| llms.txt | A community-proposed convention from Jeremy Howard (Answer.AI): a Markdown index of key resources at the site root; not a formal standard |
| ai.txt | A proposed extension of robots.txt semantics for AI crawler permissions, with control by purpose (training vs retrieval); still a draft with minimal adoption |
| Crawl budget | How many URLs a crawler fetches from a domain in a given window; pages over budget may miss AI citation pools in time |
| Index | In AI search, the data store an answer engine searches during retrieval, often vector embeddings plus metadata instead of only keyword-inverted lists |
Content structure for AI
Schema pages get cited two to three times more often than equivalent pages without markup, so this group pays for itself fastest. If I could add only one type, it would be FAQ schema, because it matches the page to the way questions actually arrive.
| Term | Meaning |
|---|---|
| Schema markup | Structured data annotating content with Schema.org vocabulary; pages with schema are cited two to three times more often by AI engines than equivalent unstructured pages |
| JSON-LD | The structured data format Google recommends for embedding Schema.org annotations, placed inside a <script type="application/ld+json"> tag |
| FAQ schema | Schema.org Question/Answer markup; the single highest-impact schema type for AI citation in 2026, since it aligns page structure with the query format |
| HowTo schema | Marks up step-by-step instructions, tagging each step with name, description and image so AI engines extract the process cleanly |
| DefinedTerm schema | Schema.org type for glossary entries and definitions; AI systems treat it as a strong signal of a reliable definitional source |
| Entity | A distinct, identifiable thing (person, organization, product, concept) whose properties and relationships can be structured; pages describing a known entity get retrieved more often |
| Ontology | A machine-readable spec of a domain's concepts and their relationships; Schema.org is a shared simplified one, and enterprises layer domain-specific ones on top |
| Structured data | Content formatted so machines parse its meaning without inference, an explicit contract between publisher and retrieval system about what a page contains |
| Named-entity recognition (NER) | An NLP technique identifying and categorizing entities in unstructured text; retrievers use it to link page content to knowledge-graph entities |
| Semantic HTML | HTML5 elements like <article>, <section>, <header> and <figure> that carry meaning about content role; AI crawlers use them to find the main body without full visual rendering |
How often does ChatGPT mention your brand?
Most founders have no idea. The answer might surprise you.
Citation metrics
Two of these get mixed up constantly: share of voice compares you against competitors on a fixed prompt set, while citation rate only asks how often you showed up at all. Most of the rest covers what can go wrong between a source and the answer.
| Term | Meaning |
|---|---|
| Share of voice (SOV) | Your brand's citations as a percentage of all brand citations across a fixed prompt set on one engine: (your citations / total category citations) x 100 |
| Citation rate | Share of tested queries where an AI engine cites your domain at least once; strong B2B benchmarks target 10-15% on category queries, and leaders exceed 30% |
| Brand mention frequency | Average number of times a brand name appears in AI responses across a prompt set, capturing ambient presence even without a hyperlinked citation |
| Source authority | An AI engine's internal read of a domain's credibility as a citation candidate, informed by backlinks, EEAT signals, structured data and historical citation patterns |
| Retrieval score | The similarity value given to a candidate document during retrieval, typically cosine similarity between query and document embeddings; higher scores enter the answer context window |
| Citation graph | A directed network of documents (nodes) and citation relationships (edges), used to find authoritative hubs and measure citation flow across a content ecosystem |
| Provenance | The documented chain of sources behind a specific AI claim; poor provenance tracking is a primary cause of attribution hallucinations |
| Attribution | Linking a claim in an AI answer back to its source document, which needs correct retrieval and correct passage-to-claim mapping |
| Hallucination | Plausible-sounding AI output that is factually wrong, unsupported by retrieved sources or wrongly attributed; researchers distinguish factual, grounding, citation and reasoning subtypes |
| Groundedness | How far an answer is supported by retrieved documents instead of inferred from model weights; a fully grounded response maps every claim to a specific retrieved passage |
Business and buyer terms
| Term | Meaning |
|---|---|
| B2B AEO | Answer engine optimization for business-to-business sales, where buyers use AI assistants to research vendors and build shortlists before contacting sales |
| Enterprise LLM adoption | Organizations integrating LLM tools into workflows, from internal knowledge management to procurement research; as of 2026, 94% of B2B buyers report using LLMs for vendor research |
| MCP (Model Context Protocol) | Open standard developed by Anthropic and donated to the Linux Foundation's Agentic AI Foundation in December 2025, defining how AI agents connect to external tools and data sources at runtime |
| Agentic buyer | A business purchaser who delegates early-stage vendor research, shortlisting or RFP drafting to an AI agent instead of doing it manually |
| AI-first buyer journey | A purchase that starts with an AI assistant prompt instead of a search query or analyst report; time-to-consideration is 40-60% shorter than traditional cycles |
| LLM procurement | LLM assistants used in a formal purchasing process, from needs analysis and vendor discovery through shortlisting and price comparison |
| Zero-click marketing | Brand content built to deliver value inside AI answers even when users never visit your site; the success metric shifts from referral traffic to citation rate |
| GEO spend | The share of a marketing budget allocated to generative engine optimization (content creation, schema implementation, authority outreach and AI citation monitoring), distinct from traditional SEO spend |
This is the vocabulary for a budget conversation, and with 94% of B2B buyers researching vendors through LLMs and a 40-60% shorter time-to-consideration on AI-first journeys, a GEO line is easy to defend.
Related SEO concepts
None of these are new, but they're the machinery under everything above. Embeddings are why a page can be retrieved for a question it never phrases word for word, and EEAT-style signals help decide which retrieved page ends up cited.
| Term | Meaning |
|---|---|
| EEAT (Experience, Expertise, Authoritativeness, Trust) | Google's four-part quality framework from the Search Quality Evaluator Guidelines; AI engines use EEAT-adjacent signals to rank citation candidates |
| Helpful Content | Google's 2022 content quality standard rewarding content written for people instead of search engines; AI engines apply a similar filter on direct-answer relevance |
| Semantic search | Matching queries to documents by conceptual meaning instead of exact keyword overlap; the foundation all major AI answer engines build on |
| Embedding | A dense numerical vector for text, generated so semantically similar content sits close together in high-dimensional space; queries and documents both get embedded for retrieval |
| Vector database | A data store optimized for embeddings, enabling fast approximate nearest-neighbor search across millions of documents; the retrieval backend most AI search engines run on |
15 hours a week manually. Or 15 minutes with RankControl.
Track citations, monitor competitors, and fix content gaps across every AI search engine. Automatically.
The vocabulary keeps growing because the surface area does, and I'd expect next quarter to bring three more crawler names and two more measurement metrics, plus at least one new acronym for an idea somebody already named. Keep the sheet handy and your next AEO meeting should take half the time. For the vendor-by-vendor bot list that pairs with the crawler section, see the complete AI crawler list.
