Someone on your team said "we're losing citation share to a competitor," and half the room nodded while the other half quietly wondered what a citation even is. That gap is normal. AI search arrived fast, and the vocabulary arrived faster than the shared understanding of it.
This is your AEO glossary: one place to learn the AI search terms that keep showing up in strategy decks, vendor demos, and the panicked Slack message after leadership asks "what's our AI search strategy?" You do not need to memorize all of it today. You need a working vocabulary you can actually use in a meeting, and a mental map of how the pieces connect.
Here is the good news. Most of answer engine optimization terminology clusters into a handful of themes: the surfaces where answers appear, the disciplines that optimize for them, the signals inside an answer, the machinery underneath, the on-page moves you control, and the metrics that make all of it legible. Learn the themes and the individual terms stop feeling like alphabet soup.
Each entry below is written as a self-contained definition block: the term, a one-line answer, then a short expansion. Quote any single block on its own. Let's build the vocabulary one cluster at a time.
Start with the surfaces where answers appear
Before you can optimize for AI search, name the places your buyers actually see answers. This is where most confusion starts, because "search" now means several different products.
Search engine. A system that returns a ranked list of links for a query. Google and Bing classic search live here. The user reads the links and composes their own answer.
Answer engine. Any AI product that returns a synthesized answer instead of a list of links. ChatGPT, Perplexity, Google AI Mode, and voice assistants qualify. The defining output is prose with inline or paneled citations, plus follow-up prompts. This shift, from ranking links to composing answers, is the whole reason AEO exists.
Generative engine. An AI system whose primary output is original synthesized text, not selected existing content. ChatGPT, Claude, Gemini, and Perplexity are all generative engines, and the term shows up in the phrase "generative engine optimization."
AI Overview (Google). The block of AI-generated summary text with source links that Google places above the traditional organic results for some queries. It differs from a featured snippet because it is multi-source and synthesized, not a single passage pulled from one page.
AI Mode (Google). A dedicated Search tab that gives expanded, multi-step AI answers instead of ten blue links. It splits your question into multiple sub-searches, then returns a synthesized response with inline links and a side panel of sources.
ChatGPT, Perplexity, Gemini, Claude. The four assistant families marketers track most. They behave differently. Only a minority of ChatGPT conversations ever trigger a web search, so head-of-conversation questions matter most. Perplexity reads roughly ten sources per query and cites only three or four of them. Gemini leans on Google Search for its grounding. Claude's citation mechanics are the least publicly documented, so treat claims about it carefully.
Conversational search and voice search. Natural-language queries that produce synthesized, multi-turn answers, whether typed or spoken. Voice was the original answer-engine surface, and it never returns a list of links to read aloud.
The takeaway: you are not optimizing for "AI." You are optimizing for specific surfaces, each with its own behavior. Many of the AI search terms below only make sense once you know which surface they describe. Pick the engines your buyers use and start there.
Lock down the umbrella terms
Here is a question that trips up smart teams: are SEO, AEO, GEO, and LLMO four different things or four names for the same thing? Mostly the second. Getting this straight saves you a lot of circular debate.
SEO (Search Engine Optimization). The original discipline: structure pages and sites so search engines rank them highly in classic blue-link results. SEO is not dead. It is the discoverability substrate that everything else builds on.
AEO (Answer Engine Optimization). The practice of structuring and formatting content so AI tools can understand, trust, and cite it as a direct answer. AEO is about becoming the cited source inside an AI response rather than the top link in a results page. It centers on clarity, extractability, and putting the answer up front.
GEO (Generative Engine Optimization). Optimizing content to be cited inside LLM-generated responses. In practice most people use GEO and AEO interchangeably. Some reserve GEO for the "LLM as recommender" framing, but you can read it as a synonym by default.
LLMO (LLM Optimization). Another synonym for the same discipline, more common in Europe and in early academic papers. Treat it as interchangeable unless a source says otherwise.
The clean mental model: SEO is the legacy discipline of ranking links. AEO, GEO, and LLMO are the disciplines of getting cited in AI answers, and they are largely the same practice under different labels. Mastering one requires the others, which is exactly why a shared aeo glossary helps your whole team stop talking past each other. Most answer engine optimization terminology is just these disciplines applied to a specific surface, signal, or metric.
Separate a citation from a mention from position zero
If you only sort out one distinction from this whole aeo glossary, make it this one. The difference between a citation and a mention decides how you measure everything.
Brand mention (in AI search). Any time an AI engine names your brand inside an answer. A mention is unstructured text. It can sit in any sentence, it can list you next to competitors, and it does not link back to your site. A mention proves awareness.
Citation (in AI search). A clickable or attributed link in an AI answer that points to a specific page on your domain as the source for a claim. The attribution is what separates a citation from a mention. A citation proves authority, because the engine is effectively saying "this brand is the source."
So what is a citation vs mention in plain terms? A mention says people know you exist. A citation says the engine trusts your page enough to point at it. A brand with many mentions and few citations is known but not trusted. The reverse is trusted but not seen. You want both, and you want to track them separately.
Common mistake: counting mentions and citations as one number. Most reporting tools bundle them together, which hides the exact gap you need to see. If your mention rate is healthy but your citation rate is flat, the fix is not more awareness, it is more citable structure on your pages. Separate the two metrics and report the gap out loud.
Source attribution. The general practice of crediting an AI-generated claim with a clickable link. Source attribution is what turns a mention into a citation. Without it, the model is speaking for itself.
Position zero. Originally the featured-snippet box above the first organic result. In the AI era the term drifted. Some people still mean the literal snippet slot, and others now mean the AI-generated answer block that sits above organic results, with your brand citation embedded inside it. This is a frequent point of confusion, so define which one you mean when you say it.
Featured snippet. A single-source answer Google pulls into a special box above the organic results, as a paragraph, list, table, or video. It differs from an AI Overview because it is one source pulled verbatim, not a multi-source synthesis.
Rich snippet. An enriched normal result that adds ratings, prices, or FAQ dropdowns to a standard blue link. Different from a featured snippet: rich snippets live in the normal result slots, featured snippets appear above them.
Share of voice and share of citation. Two relative metrics. Share of voice is your percentage of total brand mentions in AI answers for a set of prompts, versus your named competitors. Share of citation is the same idea using citations instead of mentions. They can tell very different stories, so report both and label which one you mean.
Decode what happens under the hood
You do not need a machine-learning degree, but a few mechanics terms make the rest of AEO click. This is the cluster that explains why front-loaded, well-structured pages win.
RAG (Retrieval-Augmented Generation). The architecture most answer engines run on. The engine retrieves relevant documents, adds them to the model's prompt, and generates an answer grounded in that retrieved content rather than the model's memory alone. RAG is why fresh, well-structured pages can influence an AI answer at all.
Grounding. Anchoring a model's response in real, retrievable source material so claims trace back to specific information. Grounding is what RAG produces, and it is the conceptual opposite of hallucination.
Hallucination. An AI response that presents false or fabricated information as fact. Rates vary enormously between systems and tasks. The point for marketers: grounding reduces it, and a well-sourced page is easier to ground against.
Query fan-out. The technique where an engine takes your single prompt, breaks it into multiple parallel sub-queries, searches all of them, and synthesizes the result. A single question commonly fans out into eight to twelve sub-queries. This is the single most important mechanic in this glossary, because most brands optimize for the head term and never show up for the sub-queries the engine actually runs. Analysis of large URL sets suggests the vast majority of brands miss citations at this fan-out level.
So the position zero fan-out meaning that people paste into a search bar is really two ideas colliding: position zero is where the answer appears, and query fan-out is how the engine assembles it. Understand both and you stop optimizing for one keyword when the engine is asking a dozen questions.
Retrieval, ranking, and chunking. Retrieval is the step where the engine looks up relevant passages by meaning rather than exact keyword. Ranking is the step where candidate passages get scored and filtered down to the few that get cited. Chunking is how your content gets broken into small, liftable units in the first place. Clear headings and short paragraphs improve all three.
Embeddings and semantic search. Embeddings are numerical representations of text where similar meanings sit close together. Semantic search matches queries to content by meaning using those embeddings. Together they are why "index meaning, not strings" describes modern retrieval.
Learn the on-page signals engines reward
Here is the reassuring part. Roughly a third of AEO lives on your own pages, in signals you fully control. Master this cluster and you have real leverage without waiting on anyone.
Entity. A recognizable, disambiguated thing: a brand, person, product, or concept. Entities are how an engine knows "Apple the company" differs from "apple the fruit." Consistent naming and structured data strengthen your entity.
Schema markup (structured data). Machine-readable JSON-LD that labels your content by type, like Article, Product, Organization, FAQ, or HowTo. AI crawlers process schema at crawl time. The strongest types for AI search, by current practitioner consensus, are FAQ, HowTo, Article, Organization, and Product.
BLUF (Bottom Line Up Front) and atomic answers. BLUF means putting the core answer in the first sentence of a section. An atomic answer is a self-contained one-to-three-sentence block that resolves a single sub-question. Both make your content easy to lift intact, which is exactly what engines do.
Pro tip: front-load the answer. Studies of top-cited pages find that around nine in ten place the answer within the first hundred words. If your page opens with a long windup and buries the answer under a heading, it is structurally hard to cite. Write the answer first, then the supporting context. This one habit moves the needle more than almost anything else on the page.
Extractability. How easily an engine can lift a fully-formed answer from your page without needing the surrounding context. Crisp answers under clear headings are extractable. The same facts buried in narrative are not.
Multi-question coverage and hub-and-spoke. Multi-question coverage means answering adjacent sub-questions on one page so it can satisfy a fan-out of related queries. Hub-and-spoke is the architecture where one pillar page owns the parent topic and spoke pages own the narrower questions. This pattern maps almost perfectly onto how engines fan queries out.
E-E-A-T. Google's framework for Experience, Expertise, Authoritativeness, and Trustworthiness. In AI search it acts as a trust filter on whether an engine considers a source credible enough to cite.
Recency signals and information gain. Recency signals are cues that your content is current: visible update dates, dated references, current year in headings. Information gain is the unique value you add beyond the consensus. Fresh pages with original insight get cited more often than restatements of what everyone already published.
llms.txt. A proposed Markdown file at /llms.txt that gives LLMs summary information about your site, conceptually similar to robots.txt but aimed at inference time. Adoption is growing but engine support is uneven, so treat it as an emerging best practice, not a guaranteed lever.
Build the metrics set that makes it legible
You cannot manage what you cannot see, and a single brand-level number tells you almost nothing. This cluster is the reporting language you will use with leadership.
Mention rate. The percentage of answers in your tracked prompt set that name your brand in any form. It captures presence in the narrative without requiring a link.
Citation rate. The percentage of answers in your tracked set that link to your domain as a source. Alongside mention rate, this is the headline AEO performance metric.
Sentiment. Whether the AI describes your brand in favorable, neutral, or unfavorable terms. "Described correctly and positively" is the reporting frame you want.
Accuracy score. Whether the engine gets your brand's facts right, or in other words how often it gives a non-hallucinated answer about you.
Visibility trend. The period-over-period change in your metrics. Slope matters more than the absolute level. A low but steeply rising number is a better position than a high but flat one.
The right starter set is small: track share of voice, citation rate, and mention rate, segmented by prompt and by engine, on a regular cadence. A single number for the whole brand is not actionable.
This is the point where a platform earns its keep, because collecting per-prompt, per-engine data by hand across ChatGPT, Perplexity, Gemini, and more gets impossible fast. This is the work DeepSmith is built for: its AI search visibility module reports mention rate, citation rate, and share of voice, with a per-platform breakdown and a competitor leaderboard, so you can see not just your numbers but who is winning your prompts and on which pages. You define what to track; it checks on a schedule and reports back.
Turn the vocabulary into prompts you actually track
A glossary is only useful if it changes what you do on Monday. The bridge from vocabulary to action is the prompt.
Prompt (in AI search). The user's question, expressed in natural language, submitted to an answer engine. Unlike a keyword, a prompt is conversational and often multi-turn. In measurement, a prompt is the thing you track. In production, it is the question your content must answer.
Prompt set (tracked prompts). The curated list of natural-language questions you monitor for AI search presence. A manual workflow might track five to ten prompts; a platform-based one tracks dozens to a few hundred, organized by buyer stage and persona. This is the cornerstone of any measurement program.
Keyword (in the AI era). Still useful as a unit of organization, but less central than in classic SEO, because engines index meaning rather than exact strings. Use keywords to seed coverage, not to anchor every heading.
Here is the move. Build your prompt set from real buyer questions, the ones from sales calls, support tickets, and the "People Also Ask" box, not from the internal names your team uses for your product. Then check how AI answers each one, find where you are invisible, and write the content that fills the gap.
That last step is where the vocabulary pays off. A citation gap is a page that ranks in classic search but never gets cited in AI answers, and closing it means restructuring for extraction: the answer up front, atomic answer blocks, schema, clean internal links. That is genuinely time-consuming to do by hand for every page, which is why teams reach for a production system.
Ready to turn the glossary into a strategy?
You now have the shared language your team was missing. The surfaces, the umbrella disciplines, the citation-versus-mention distinction, the machinery, the on-page signals, and the metrics. That is more of the AI search terms in play than most marketing teams can define, and you got here in one read.
The next step is small: pick three to five real buyer prompts, search them in the engines your audience uses, and note where you show up and where you do not. That single exercise turns this glossary from theory into a to-do list.
When you are ready to do it at scale, that is what DeepSmith was built for. It tracks your visibility across ChatGPT, Perplexity, Gemini, Claude, and Google AI Mode, finds the prompts where you are invisible or losing, and produces publish-ready, on-brand articles to close those gaps, grounded in your stored brand context so the output actually sounds like you. Start a free DeepSmith trial and see your real citation gaps before you write another word.
