AI Search Glossary.
Every new discipline arrives with its own vocabulary, and AI search has produced more of it in two years than SEO managed in twenty. AEO, GEO, LLMO, AIO: the acronyms overlap, the definitions drift depending on who is selling what, and brands end up buying services they cannot describe internally.
This glossary is our attempt to settle the language. Each entry below is written the way we would explain it to a client on a call, with no hedging, no borrowed jargon, and no definition longer than it needs to be. The terms are grouped by the job they do rather than listed alphabetically, so you can read a section end to end and come away understanding a whole area of the work.
Where a term connects to something we do, we have linked the relevant service page. Where it connects to a public standard or piece of platform documentation, we have linked the primary source instead of a summary of it.
1. The Foundations
AI Search
The broad shift from typing keywords into a results page to asking a question and receiving a written answer. The commercial consequence is simple. There are fewer links to click and fewer brands named, so the cost of being left out has gone up sharply.
Answer Engine
Any system that responds to a question with a composed answer instead of a ranked list of pages. ChatGPT, Perplexity, Gemini, Copilot, and Google's AI Overviews all qualify, even though they source and cite information very differently.
Answer Engine Optimization (AEO)
The work of making a business the answer rather than a result, structuring pages, facts, and positioning so an answer engine selects your brand when it composes a response. See our AEO services.
Generative Engine Optimization (GEO
Optimization aimed specifically at systems that generate text on the fly by pulling from retrieved sources. GEO leans heavily on passage-level clarity, because the model has to be able to lift a self-contained chunk of your page into its answer without editing it.
Large Language Model Optimization (LLMO)
Influencing what a model believes about your brand, not just what it can retrieve about you in the moment. LLMO is the long game. It concerns training data, widely repeated descriptions, and the version of your company that shows up when nothing is retrieved at all.
Artificial Intelligence Optimization (AIO)
An umbrella label covering every technique that improves how AI systems read, classify, and recommend a business. Useful as a category heading, less useful as a scope of work.
AI SEO
Traditional search optimization carried forward into AI surfaces. The technical fundamentals still apply, since a page that cannot be crawled or indexed cannot be cited, but the objective moves from position to selection.
Large Language Model (LLM)
The class of model behind ChatGPT, Claude, Gemini, and Copilot. It composes answers by predicting language from patterns absorbed across enormous volumes of text, which is why what the web repeats about you matters as much as what your own site says.
Retrieval-Augmented Generation (RAG)
An architecture where the model first fetches relevant documents, then writes an answer grounded in them. Nearly every AI search product uses some form of retrieval, which is precisely why indexation and crawler access remain non-negotiable.
Training Data vs. Retrieval
Two separate routes into an AI answer. Training data is what the model absorbed before release and cannot be changed after the fact. Retrieval is what it fetches live and can be influenced this quarter. Most practical wins come from the second, most durable wins from the first.
Knowledge Cutoff
The date beyond which a model's built-in knowledge stops. Anything about your brand that happened after that date only reaches the model through retrieval, which is one reason recent rebrands so often produce outdated AI descriptions.
2. Getting Named, Cited, and Recommended
Citation
An explicit, usually linked reference to your page inside an AI answer. This is the highest-value outcome in AI search, because it carries attribution, referral traffic, and a visible credibility signal to the reader.
Mention
Your brand name appearing in an answer without a link attached. Worth less in traffic terms than a citation, but a genuine indicator that the model associates your business with the topic.
Recommendation
The model does not simply describe you. It puts you forward as the right choice. Recommendations cluster around commercial queries such as "best," "top," and "who should I use for," which makes them the most contested and most valuable outcome of all.
Inclusion
Making it into the answer in any form. AI search has no second page. You are in the response or you do not exist within it, which is a harsher distribution than the ten blue links ever were.
Exclusion
Being absent from answers where your competitors appear. Rarely the result of one broken thing. It is usually a stack of small ambiguities the model resolves by naming someone clearer.
Zero-Click Answer
A response complete enough that the user never visits a source. Frustrating for traffic reporting, but a mention inside one still shapes what the buyer believes before they ever reach your site.
Citation Frequency
How often your brand surfaces across a defined set of tracked prompts. The core volume metric in AI search reporting, and the one that shows movement first.
Share of Voice (AI)
Your slice of all brand mentions across a tracked prompt set, measured against named competitors. It answers the question clients actually ask, which is not "are we visible" but "are we more visible than them."
Answer Position
Where you land within a response: named first, buried in a list of nine, or mentioned only in a caveat. Position tends to track buyer preference more closely than raw mention counts do.
Sentiment Accuracy
Whether the model's characterization of your brand is both positive and correct. A flattering description built on outdated facts is still a problem worth fixing.
Passage-Level Citability
How cleanly a single paragraph can be extracted and reused without surrounding context. Self-contained passages get quoted. Paragraphs that depend on the three above them get skipped.
3. Your Brand as an Entity
Entity
A specific, identifiable thing, such as your company, your founder, or your product line, that AI systems reason about as an object with attributes and relationships. Models do not match keywords anymore. They resolve entities.
Entity Resolution
The process of deciding which real-world thing a name refers to. If three companies share your trading name, resolution is the step where you either get correctly identified or quietly merged with someone else.
Entity Clarity
How consistently and unambiguously your business is described across every source a model can reach. The practical test: can AI state what you do, for whom, and where, in one sentence, without contradicting itself?
Disambiguation
The signals that separate you from similarly named businesses, including registration details, addresses, founder names, and verified profiles. Weak disambiguation is one of the most common causes of a brand simply vanishing from answers.
Knowledge Graph
A structured map of entities and how they relate to one another. Google's version feeds Gemini and AI Overviews directly, which makes a presence in it disproportionately valuable.
Wikidata
An open, structured knowledge base that flows into search and AI systems. A properly sourced entry is one of the few entity assets you can build deliberately rather than wait for.
Brand Description Consistency
Using the same core sentence about your business everywhere, across your site, directories, profiles, press, and partner pages. Every variation gives the model another possible interpretation to weigh, and uncertainty suppresses recommendation.
NAP Consistency
Identical name, address, and phone number across every listing. Minor formatting drift, such as "Ave" versus "Avenue," is enough to fracture a single entity into several partial ones.
Interpretive Drift
Gradual change in how AI describes your business, usually after a repositioning, a rebrand, or a run of off-topic content. Left unattended, drift becomes the new default description.
4. Machine-Readable Signals
Structured Data
Code that states plainly what a page and a business are, rather than leaving a model to infer it from prose. It converts your claims into facts a system can extract without ambiguity.
Schema.org: The shared vocabulary used to express structured data, supported across major search and AI platforms. See Schema.org.
JSON-LD: The preferred format for adding schema markup. It is a self-contained script block that sits apart from your visible content and is easy to maintain. See Google’s structured data documentation.
Organization Schema: Markup that defines the business itself, covering legal name, logo, contact points, and the profiles it owns. This is the anchor every other entity signal attaches to. See schema.org/Organization.
sameAs: A schema property listing the official profiles that belong to the same entity, tying your site, socials, and directory listings into one identity. See schema.org/sameAs.
FAQ Page Schema
Markup that pairs questions with answers in a machine-readable format. Useful precisely because it mirrors how answer engines already think about content.
AI Crawler
The bots that fetch pages on behalf of AI platforms, including GPTBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot, and Google-Extended. Blocking them, deliberately or accidentally, removes you from retrieval entirely.
robots.txt
The file that grants or denies crawler access. A single overzealous rule here can undo an entire visibility program, and it is one of the first things we check.
llms.txt
An emerging plain-text file that gives AI systems a curated summary of a site and its most important pages. Support is still uneven, but the cost of publishing one is close to zero.
Crawlability
Whether bots can actually reach your content. Login walls, heavy client-side rendering, and broken internal paths all quietly reduce it.
Indexability
Whether a crawled page is eligible to be stored and retrieved later. Content that never enters an index cannot be pulled into an answer no matter how good it is.
JavaScript Rendering
Whether your content exists in the raw HTML or only appears after scripts run. Several AI crawlers are far less patient than Googlebot, so content that renders late may never be seen.
Content Chunking
Structuring a page into discrete, self-explanatory blocks, with clear headings, direct opening sentences, and one idea per section. It is the single most reliable formatting change for improving citation rates.
Semantic HTML
Using markup that describes meaning rather than appearance, including real headings, real lists, and real tables. It gives parsers the outline of your argument for free.
5. How People Actually Query AI
Prompt
The question or instruction typed into an AI system. Prompts run far longer and more conversational than search queries, which is why keyword lists alone no longer describe demand.
Prompt Set
The defined group of questions we track for a client, usually a mix of commercial, comparative, and problem-led phrasings. It is the measurement baseline everything else is reported against.
Comparative Prompt
A request to weigh providers against each other, such as "best AI search agency in Singapore." These are the highest-intent prompts in most categories and the ones competitors fight hardest over.
Navigational Prompt
A question about your brand by name. The fastest diagnostic available: ask the model who you are, and its answer tells you exactly how clear your entity signals are.
Problem-Led Prompt
A question framed around a symptom rather than a solution, such as "why isn't my site showing up in ChatGPT." Won by demonstrated expertise rather than by category positioning.
Query Fan-Out
The tendency of AI systems to silently expand one user question into several background searches before answering. It means you can be retrieved for phrasings the user never typed.
Prompt Volatility
The way near-identical questions can return meaningfully different brand sets. Any credible reporting accounts for this by sampling repeatedly rather than screenshotting once.
Hallucination
A model stating something false with complete confidence. When it happens to a brand, the root cause is usually a gap in verifiable information that the model filled by inference.
Grounding
Anchoring an answer to retrieved, checkable sources rather than model memory. Grounded answers are where citations live, which makes retrievability a commercial concern rather than a technical one.
6. Trust, Proof, and Off-Site Signals
Ecosystem Validation
Independent sources across the web describing your business the same way. When third parties agree with your own site, model confidence rises. When they disagree, it collapses.
AI Link Building
Earning references from the sources AI systems actually retrieve and trust, rather than chasing volume for its own sake. The target is influence over retrieval, not a metric on a dashboard.
Community Signal
Mentions in forums, Q&A threads, and discussion platforms, Reddit above all, which several major models retrieve from heavily when handling recommendation queries. See our Reddit management services.
Unlinked Brand Mention
Your name appearing in third-party content with no hyperlink. Traditional SEO discounted these. Language models do not, because the name itself carries the association.
E-E-A-T
Experience, Expertise, Authoritativeness, and Trustworthiness. It is Google's quality framework, and a serviceable checklist for AI credibility more broadly. Named authors, verifiable credentials, and first-hand evidence all count.
Source Authority
How much weight a model places on a domain when deciding what to repeat. Established publications, official documentation, and recognized industry bodies carry more than a well-optimized blog.
Signal Conflict
Two credible sources telling a model different things about you, such as different service lists, locations, or founding dates. The usual resolution is to name a competitor instead.
7. The Platforms
ChatGPT
OpenAI's assistant, and the largest single source of AI-referred traffic for most brands. Its live search leans on the Bing index, which makes Bing visibility a practical prerequisite.
Perplexity
An answer engine built around explicit citation. It crawls in real time and shows its sources, so well-structured, genuinely current pages are rewarded quickly.
Google AI Overviews
The generated summary above Google's traditional results. Sourced from Google's own index and shaped by Gemini, it converts organic visibility into AI visibility more directly than any other surface.
Google AI Mode
Google's fully conversational search experience, where the answer replaces the results page rather than sitting above it. It relies more heavily on query fan-out than AI Overviews do.
Gemini
Google's assistant, drawing on Google's index and Knowledge Graph. Entity presence within Google's ecosystem carries unusual weight here.
Microsoft Copilot
Microsoft's assistant, built on Bing and connected to LinkedIn data. Underrated in most strategies, and often the easiest platform to gain ground on.
Claude
Anthropic's assistant, which cites when it searches the web and favors clear, factual, well-attributed pages. See our guide on whether Claude cites sources.
Bing
Microsoft's index, and because it sits underneath both ChatGPT search and Copilot, one of the highest-leverage places to be properly indexed. Start with Bing Webmaster Tools.
AI Agent
A system that carries out multi-step tasks on a user's behalf, including researching, comparing, and increasingly transacting. The next surface where being machine-legible determines whether you are considered at all.
Still Deciding Which of These You Actually Need?
Most brands do not need all of it. They need to know which two or three signals are holding them back, and that is a diagnosis, not a package.
We audit how the major AI platforms currently describe, cite, and rank your business, then tell you plainly what is worth fixing first.
Book a strategy call and we will walk you through what the models are saying about you today.
Frequently Asked Questions
What is the difference between AEO, GEO, and LLMO?
AEO is about being selected as the answer to a question. GEO is about being retrievable and quotable when a model composes that answer live. LLMO is about what the model believes regarding your brand even when it retrieves nothing at all. In practice the three overlap heavily, and most brands need work across all three rather than a choice between them.
Do I still need traditional SEO if I am optimizing for AI search?
Yes. Several major AI surfaces are built directly on conventional search indexes, so a page that cannot be crawled or indexed cannot be cited. Technical SEO has not been replaced. It has become the entry requirement rather than the strategy.
How do I know if AI platforms are describing my business correctly?
Ask them. Query each major platform by your brand name and compare the answers against your own positioning. Wrong service lists, outdated locations, and confusion with similarly named companies are all common, and all fixable once identified.
Which AI platform should I prioritize?
It depends on where your buyers are. B2B audiences skew toward ChatGPT and Copilot, research-heavy buyers toward Perplexity, and broad consumer discovery still runs through Google AI Overviews. We recommend prioritizing based on your tracked prompt data rather than on platform market share.
How often does this glossary change?
We review it as platforms shift, which in this field means several times a year. Terms in current use get updated definitions, and terms that stop meaning anything useful get removed.