Reference · Free
This is a living AEO glossary: 62 plain-language definitions of Answer Engine Optimization, GEO, and AI search terms you can cite — not a ranking or citation guarantee.
It covers concepts from llms.txt and RAG to citation tracking for marketers, developers, and anyone who needs a clear definition. Every entry names the concept in the first sentence and stays short enough to stand alone in an AI answer. Where a deeper guide or free tool exists on this site, the card links through. English and Turkish share one data source so wording and schema never drift apart.
Answer Engine Optimization (AEO) is the practice of structuring web content so AI answer systems such as ChatGPT, Google AI Overviews, Perplexity, and Gemini can retrieve, understand, and cite it accurately. AEO combines answer-first writing, clear entities, structured data, and crawlable pages; it does not guarantee rankings or citations.
Generative Engine Optimization (GEO) is the discipline of making a brand the cited or recommended source across AI-generated answers. Where AEO optimizes a single page, GEO works at brand level: entity authority, third-party mentions, knowledge-graph presence, and consistent facts about the brand across the web.
AI search optimization is the umbrella term for improving visibility in AI-assisted search surfaces, covering both page-level AEO and brand-level GEO. It targets systems that synthesize answers from multiple sources — AI Overviews, ChatGPT Search, Perplexity, Copilot — rather than ranked lists of links.
LLM Optimization (LLMO) is an alternative label for optimizing content and brand signals so large language models mention or cite them. In practice LLMO overlaps almost entirely with AEO and GEO; the term is used mainly in marketing literature rather than as a distinct technical discipline.
An answer engine is a search system that returns a synthesized, direct answer instead of a ranked list of links. Answer engines — ChatGPT Search, Perplexity, Google AI Overviews, Bing Copilot — retrieve eligible pages, rank passages by relevance and trust, then generate one response with optional citations.
Google AI Overviews are AI-generated summaries shown above classic results for eligible queries. They draw on indexed pages that meet baseline quality, favouring extractable, well-structured, trustworthy passages. Strong SEO foundations plus AEO-style schema and answer-first formatting improve eligibility for both snippets and Overviews.
Google AI Mode is a conversational search experience in which Gemini runs multi-step, agentic research across the index and answers follow-up questions in one thread. AI Mode favours sources it can decompose into clear, factual passages, making entity clarity and structured content even more important than in classic search.
A zero-click search is a query resolved entirely on the results page — through a featured snippet, knowledge panel, or AI Overview — without the user visiting any website. As AI answers grow, zero-click share rises, shifting the goal from clicks alone to being the cited or named source.
A featured snippet is a highlighted answer box Google shows above organic results, extracted from a page that answers the query directly. Featured snippets reward the same patterns AEO does — concise definitions, question headings, lists, and tables — and often feed the passages AI Overviews reuse.
Conversational search is querying in natural language across multiple turns, where the system keeps context and refines answers. It changes optimization targets: long, specific questions replace short keywords, and content that anticipates follow-up questions — FAQ blocks, comparison sections — earns more selections than keyword-dense copy.
Voice search optimization prepares content to be read aloud by assistants such as Google Assistant, Siri, and Alexa. It relies on concise spoken-length answers, question-format headings, and speakable structured data marking which passages are suitable for text-to-speech — one of the earliest practical forms of AEO.
Answer-first content opens with a direct, complete answer — typically in the first 40–60 words — before adding context, evidence, and nuance. The pattern mirrors how answer engines extract passages: a self-contained lead paragraph is easier to quote accurately than an answer buried mid-article.
A large language model (LLM) is an AI system trained on massive text corpora to predict and generate language. LLMs — GPT, Gemini, Claude, Llama — power answer engines and chat assistants. For visibility work, what matters is which sources an LLM saw in training and which it retrieves live.
Retrieval-augmented generation (RAG) is the architecture behind most AI search: the system first retrieves relevant documents or passages, then generates an answer grounded in that retrieved material. RAG is why crawlability, extractable structure, and passage-level clarity directly determine whether a page can be cited.
Grounding is anchoring an AI-generated answer in verifiable retrieved sources rather than the model's internal memory alone. Grounded systems quote or cite documents fetched at answer time, which reduces hallucination. Pages that state facts clearly, with named entities and dates, are easier to ground against.
An AI citation is a visible source reference inside an AI-generated answer — a link, footnote, or named attribution pointing to the page the claim came from. Citations are the primary currency of AI search visibility: they drive referral traffic, brand exposure, and trust even when users do not click.
An AI hallucination is a confident but false statement generated by a language model — an invented fact, price, feature, or quote. For brands, hallucinations become reputation risk when assistants repeat wrong claims. The main defence is publishing clear, consistent, well-structured facts the models can retrieve and ground against.
Training data is what a model learned before its knowledge cutoff; live retrieval is what it fetches from the web at answer time. AI visibility has two clocks: long-term presence in training corpora (mentions, authority, longevity) and short-term retrievability (crawlable, extractable, fresh pages).
A knowledge cutoff is the date after which a language model has no training data. Anything published later is invisible to the model unless it retrieves the page live. This is why time-sensitive content must be crawlable by AI bots — retrieval is the only path into post-cutoff answers.
A context window is the maximum amount of text a language model can process in one request, measured in tokens. Retrieval systems fill it with the most relevant passages, not whole sites — so each section of a page must stand alone, because models often see fragments, not full documents.
An embedding is a numeric vector representing the meaning of a piece of text, produced by an AI model. Search systems compare embeddings to find semantically similar passages regardless of exact wording — the mechanism behind semantic search and why covering a topic clearly matters more than repeating keywords.
Semantic search matches queries to content by meaning rather than exact keywords, typically by comparing embedding vectors. A page can surface for phrasings it never uses if it covers the concept well — shifting optimization from keyword density to complete, unambiguous treatment of a topic and its entities.
Chunking is how retrieval systems split pages into passage-sized pieces before indexing and ranking them. Chunk boundaries usually follow headings and paragraphs, so clean heading hierarchy and self-contained sections determine whether an extracted chunk still makes sense — and can be cited — out of context.
Passage ranking scores individual sections of a page — not just the whole document — against a query. Answer engines select the best-scoring passages to ground their response. A single well-structured section can earn a citation even when the rest of the page targets something else entirely.
llms.txt is a proposed plain-text file placed at a site root that tells AI systems which pages matter most and how the site is organized. Unlike robots.txt, which controls access, llms.txt curates priority content for language models. Adoption is early but growing, and publishing one is a low-cost readiness signal.
robots.txt for AI bots is the practice of using user-agent rules to allow or block AI crawlers such as GPTBot, ClaudeBot, and PerplexityBot. Blocking removes a site from training and retrieval; allowing enables citations. Each engine publishes its own user-agent strings, so rules must be maintained per bot.
GPTBot is the web crawler operated by OpenAI to collect content for model training, with related agents handling live search fetches for ChatGPT. Site owners control it through robots.txt user-agent rules. Whether to allow GPTBot is a strategic choice between content protection and visibility inside ChatGPT answers.
AI crawlers are bots that fetch web content for AI systems — GPTBot and OAI-SearchBot for OpenAI, ClaudeBot for Anthropic, PerplexityBot for Perplexity, Google-Extended for Gemini training. They split into training crawlers and live-retrieval fetchers; server logs reveal which ones actually visit a site and how often.
Structured data is machine-readable markup — most commonly JSON-LD using Schema.org vocabulary — that labels the entities and facts on a page: who wrote it, what it describes, when it changed. Structured data removes guesswork for search engines and AI systems, improving eligibility for rich results and accurate citations.
JSON-LD (JavaScript Object Notation for Linked Data) is the recommended format for embedding structured data in a page: a script block, separate from visible HTML, describing entities and their relationships. Google explicitly prefers JSON-LD over microdata and RDFa because it is easier to write, validate, and maintain.
Schema.org is the shared vocabulary — founded by Google, Microsoft, Yahoo, and Yandex — that defines the types and properties used in structured data: Article, Person, Organization, FAQPage, Product, and hundreds more. It is the common language a page uses to state facts machines can rely on.
The @graph method combines all structured data on a page into one JSON-LD block where every entity has an @id and entities reference each other — the WebPage links to its author, the author to the Organization. Connected graphs resolve entities more reliably than isolated schema fragments scattered across a page.
FAQPage schema is structured data marking question-and-answer pairs on a page. Google now shows FAQ rich results only for a narrow set of authoritative sites, but the markup remains valuable for AEO: it hands answer engines pre-packaged, extractable Q&A units that match how users phrase queries.
HowTo schema is structured data describing step-by-step instructions: steps, order, tools, and estimated time. Google retired HowTo rich results in 2023, yet the markup still helps AI systems parse procedural content into clean, ordered steps — the format assistants reuse when answering how-to questions.
Speakable markup is structured data (SpeakableSpecification) pointing to the passages of a page best suited for text-to-speech — short, self-contained summaries an assistant can read aloud. Support is limited and news-oriented, but the discipline it enforces, writing concise extractable blocks, benefits every answer surface.
A canonical URL is the single address declared as the authoritative version of a page when duplicates or variants exist. Consistent canonicals concentrate signals on one URL — critical for AI retrieval, because engines cite one address, and split signals mean weaker passages competing against each other.
Crawlability is the degree to which bots can fetch and read a page: reachable URL, permissive robots rules, fast HTML response, and content that does not require JavaScript rendering. For AI search it is the gate condition — an unreachable page cannot be retrieved, grounded against, or cited.
Extractability is how easily a machine can lift accurate, self-contained answers from a page: clear headings, answer-first paragraphs, lists, tables, and unambiguous entity references. High extractability is the page-level property AEO ultimately optimizes — crawlable pages get fetched, extractable pages get quoted.
An entity is a uniquely identifiable thing — a person, brand, product, place, or concept — that search and AI systems track independently of the words used to name it. Entities, not keywords, are the units modern systems reason about; a page ranks and gets cited for the entities it clarifies.
Entity SEO is optimizing how clearly a site defines and connects its entities: consistent naming across pages, structured data with stable identifiers, sameAs links to authoritative profiles, and unambiguous descriptions. Strong entity signals let both Google and language models attribute facts to the right brand or person.
A knowledge graph is a structured database of entities and the relationships between them — who founded what, which product belongs to which company. Google maintains the best-known one, but AI systems build similar internal representations. Presence in knowledge graphs is a core GEO objective, feeding accurate brand answers.
A knowledge panel is the information box Google shows for a recognized entity — a person, brand, or organization — populated from its Knowledge Graph. Earning one confirms Google resolved the entity confidently. Panels are influenced through structured data, consistent web-wide facts, and authoritative third-party coverage, not direct editing alone.
sameAs is a Schema.org property linking an entity in structured data to its authoritative profiles elsewhere — Wikipedia, Wikidata, LinkedIn, GitHub, official social accounts. sameAs links are the cheapest strong signal for entity resolution: they tell machines that scattered references across the web describe one identical thing.
Entity disambiguation is helping machines tell apart things that share a name — the fruit Apple versus the company, two people called the same thing. Pages disambiguate through context, structured data, and sameAs links. Ambiguous entities get merged, misattributed, or skipped in AI answers, so disambiguation protects accuracy.
Topical authority is the credibility a site earns by covering a subject completely and coherently — connected guides, definitions, tools, and answers rather than isolated posts. Both search engines and AI systems prefer sources that demonstrate depth, so authority compounds: each new piece strengthens every existing one.
Semantic HTML uses elements that describe meaning — article, nav, table, heading levels — instead of generic containers styled to look structured. Parsers and AI chunkers rely on these elements to segment pages correctly, so semantic markup directly improves how faithfully content survives extraction.
A topic cluster is a content architecture where one pillar page covers a subject broadly and linked supporting pages cover each subtopic in depth, all interlinked. Clusters concentrate topical authority and give AI systems a navigable map of expertise — this glossary and its related guides form one.
Information gain measures how much genuinely new information a page adds beyond what already exists on a topic — original data, first-hand experience, novel analysis. Systems increasingly reward gain over repetition: a page restating the consensus competes with thousands; a page adding something new becomes the source others cite.
E-E-A-T stands for Experience, Expertise, Authoritativeness, and Trustworthiness — the framework in Google quality guidelines for judging content credibility. It is not a direct ranking score but shapes systems and reviewer training. AI engines approximate the same judgment: identifiable authors, credentials, citations, and first-hand experience earn selection.
YMYL (Your Money or Your Life) describes topics that can seriously affect health, finances, safety, or wellbeing — medical advice, legal guidance, financial decisions. Both Google and AI systems apply stricter trust thresholds to YMYL content, demanding stronger authorship, sourcing, and accuracy before ranking or citing it.
Author markup is structured data identifying who wrote a page — a Person entity with name, role, credentials, and sameAs links to verifiable profiles. Machine-readable authorship converts an anonymous page into an accountable one, strengthening E-E-A-T signals and giving AI systems a named source to attribute.
Brand mentions are references to a brand anywhere on the web — linked or unlinked, in articles, forums, reviews, or directories. Language models learn brand associations largely from mentions in training data, which makes consistent, accurate, widespread mentions a core GEO input alongside classic backlinks.
Digital PR is earning coverage, mentions, and links from independent publications through newsworthy assets — original research, data studies, expert commentary, and tools. For AI visibility it does double duty: third-party coverage builds the mention footprint models learn from and supplies the authority signals retrieval systems weigh.
Freshness signals indicate how current a page is: visible last-updated dates, dateModified in structured data, updated facts, and recent crawl activity. Answer engines prefer fresh sources for time-sensitive queries, and stale pages lose citations to newer ones even when the underlying content barely changed.
Citation-worthiness is the combination of qualities that makes a passage safe and attractive for an AI system to quote: verifiable facts, named sources, precise language, clear authorship, and low ambiguity. Confident, hedge-free, specific statements get cited; vague marketing copy gets skipped regardless of ranking.
AI visibility is the degree to which a brand or page appears in AI-generated answers — cited, named, or recommended across ChatGPT, Perplexity, Gemini, and AI Overviews. Unlike rankings, it has no official position data, so it is measured by systematically asking engines questions and recording who appears.
AI share of voice is the percentage of relevant AI answers in which a brand appears compared to competitors, measured across a fixed set of prompts over time. It is the AI-search equivalent of ranking share: the primary competitive metric for GEO programs and citation tracking tools.
Citation tracking is systematically monitoring when, where, and how AI engines cite a brand or URL: which prompts trigger citations, which pages get quoted, and how answers change over time. Because engines personalize and vary responses, tracking relies on repeated sampling rather than single spot checks.
Prompt monitoring is running a curated set of realistic user questions against AI engines on a schedule and recording the answers — which brands appear, which sources get cited, what claims are made. It is the data-collection layer beneath share-of-voice, sentiment, and hallucination metrics.
An AEO score is a heuristic 0–100 summary of how ready a page is for answer engines, weighing schema validity, heading structure, answer-first writing, crawlability, and trust signals. It diagnoses on-page readiness; it is not an official metric from any platform and guarantees no rankings or citations.
Brand sentiment in AI answers is the tone with which engines describe a brand when asked — positive, neutral, negative, or mixed — and the specific claims attached. Sentiment drifts as training data and retrieved sources change, so it is tracked over time alongside citations rather than checked once.
AI referral traffic is visits arriving from AI surfaces — users clicking citations in ChatGPT, Perplexity, Copilot, or AI Overviews. It is identified in analytics through referrer domains and UTM patterns. Volumes are smaller than classic organic traffic but convert well, because users arrive pre-qualified by the answer.
How to use this glossary — answered directly.