Terms are grouped by topic. Where a chapter of this guide explains a term in depth, the entry links to it.
Foundations
AEO
Answer engine optimization: the practice of making content easy for AI answer engines to find, understand, and cite.
Why it matters
What it shows
The work side of the field: the content, structure, and sources that make a brand easier for an answer engine to use.
What it does not show
Results. AEO describes actions; AI visibility describes what those actions produced in the engines' answers.
AI assistant
A conversational AI product that answers questions, drafts text, and searches the web when asked.
Why it matters
Why it matters
Assistants are now a place where buyers research vendors and build shortlists, often before any website visit.
What it is not
A search engine in the traditional sense. Some assistants search the web live, and others answer only from what the model learned in training.
AI search
Search experiences in which an AI model reads the question, gathers sources, and writes one answer instead of returning a list of links.
Why it matters
Why it matters
Buyers ask full questions and read one written answer, so a brand is either inside that answer or absent from it. Each engine builds its answer differently, which is why results for the same question differ from engine to engine.
What it is not
A single product. The term covers Google AI Overviews and AI Mode, ChatGPT, Gemini, Perplexity, Claude, Copilot, and others. It also does not replace classic search, which still runs alongside it.
AI search intelligence
The continuous measurement of how AI engines mention, rank, and cite brands, turned into ranked actions for improving that presence.
Why it matters
Why it matters
Answers change by engine, by prompt, and from one day to the next, so a one-time check goes stale quickly. Intelligence means the measurement keeps running and every finding points to a next step, such as a page to fix, a source to earn, or a gap to close.
What it is not
A one-time audit, a content-writing service, or a traffic report. It measures the discovery layer, meaning what engines say and cite, and does not claim to measure visits, sales, or revenue.
AI visibility
How prominently a brand appears in the answers AI engines give, measured by mention rate, share of voice, average position, and citation share.
Why it matters
What it shows
How often and how prominently AI engines bring a brand into their answers to buyer questions. The four measures above describe different parts of that picture and are read side by side, not added into one number.
What it does not show
Traffic, sign-ups, or revenue. AI visibility measures what the engines say, not what visitors then do.
Answer engine
A system that responds to a question with a generated answer instead of a list of links, such as ChatGPT, Gemini, or Perplexity.
Why it matters
What it shows
The kind of system AI visibility is measured on. Each answer engine builds its answers differently and cites sources differently, so each is tracked separately.
What it does not show
A fixed list of engines. New answer engines appear and existing ones change how they answer, which is why tracking is repeated over time.
Generative AI
AI that produces new text, images, audio, or code from a prompt instead of only sorting or labeling existing data.
Why it matters
Why it matters
It is the technology behind AI answers. Because the output is generated rather than retrieved verbatim, two runs of the same prompt can differ.
What it is not
A guarantee of accuracy. Generated text can be fluent and wrong, which is why sources and verification matter.
Generative engine
An AI system that writes an answer from retrieved sources instead of listing ranked pages.
Why it matters
Why it matters
The term comes from the research that named generative engine optimization. It describes the same kind of system as an answer engine, seen from the side of how the answer is produced.
What it is not
A search index. A generative engine may use one to find sources, but its output is written text, not a ranked list.
GEO
Generative engine optimization: the practice of shaping how a brand appears in answers generated by AI engines.
Why it matters
What it shows
The same field as AEO under another name. The two terms overlap heavily, and which one a team uses is mostly a matter of habit.
What it does not show
A separate method. The comparison article sets out where the two terms differ in emphasis.
LLMO
Large language model optimization: another label for shaping how language models and the tools built on them represent a brand.
Why it matters
Why it matters
The label stresses the model itself, including what it recalls from training, where GEO and AEO stress the answer a user sees. In practice the three terms describe overlapping work.
What it is not
A distinct method with its own proven tactics. Teams that use the term still work through content, authority, technical access, and measurement.
AI search surfaces
AI Mode
Google's dedicated AI search experience, where a question is answered with a written response and can be followed up conversationally.
Why it matters
Why it matters
It handles longer, more specific questions than a typical search box, and Google describes it as able to use query fan-out to gather sources across subtopics.
What it is not
A replacement for Google Search or the same experience as AI Overviews. It is a separate surface with its own answers and links.
AI Overviews
Google's AI-written summary shown at the top of some search results, with links to supporting pages.
Why it matters
Why it matters
It sits inside Google Search, so ordinary search eligibility applies. Google states that a page must be indexed and eligible to appear in Search with a snippet to be shown as a supporting link.
What it is not
The same thing as AI Mode. Google notes that the two may use different models and techniques, so their responses and links vary.
Conversational search
Search done as a back-and-forth dialogue in full sentences instead of a string of keywords.
Why it matters
Why it matters
Follow-up questions narrow the answer step by step, so a brand can appear in the first reply, a later one, or neither.
What it is not
Voice search. Conversational search can be typed, and voice queries can still be short keyword phrases.
Supporting link
A link shown beside an AI answer as a source for the claims the answer makes.
Why it matters
Why it matters
It is the link a reader can click to check or continue from the answer, so it is where an AI answer can send a visit.
What it is not
A ranking position. Links can appear in several arrangements, and an answer can include a brand in its text without linking to it.
Search basics
Backlink
A link from one website to a page on another.
Why it matters
Why it matters
Links from independent sites have long been a signal of credibility in classic search, and the pages that earn them are often the same pages AI engines read.
What it is not
A guaranteed ranking or citation. Link quantity matters less than who is linking and why.
Branded query
A search or prompt that includes a specific brand's name.
Why it matters
Why it matters
These show how people look for a brand they already know, such as to compare it, check a price, or read reviews.
What it is not
Discovery. A buyer who types a brand name has already found it, so branded queries say little about how new buyers first meet the brand.
Featured snippet
A highlighted excerpt at the top of Google results that answers a query directly and links to its source page.
Why it matters
Why it matters
It was the earlier form of the answer-first result, and the habit of writing a clear direct answer for it carries over to AI answers.
What it is not
An AI-written answer. A snippet is lifted from a single page, while an AI answer is composed from several.
Keyword
A word or phrase a person types into a search engine.
Why it matters
Why it matters
Keyword demand shows what people look for in classic search. In AI search the unit is the prompt, which is usually longer and phrased as a full question.
What it is not
A prompt. A keyword is a short search term, and a prompt is a full instruction or question given to an AI engine.
People Also Ask
A Google results feature that lists related questions, each expanding into a short answer with a source link.
Why it matters
Why it matters
The questions it lists show how a topic breaks into related questions, which is a useful hint for the sub-questions an AI engine may also ask.
What it is not
A complete map of what AI engines search for. It reflects Google's own question data and nothing more.
Search intent
The goal behind a query, usually grouped as informational, navigational, commercial, or transactional.
Why it matters
Why it matters
The same topic can carry different intents, and an engine tends to answer each one differently. Matching intent decides which page, if any, an engine reaches for.
What it is not
The words in the query. Two queries with different words can share one intent, and one query can carry several.
Search volume
The estimated number of times a query is searched in a given period, as reported by a keyword tool.
Why it matters
Why it matters
It is a measure of real demand and is useful for choosing which topics to track. Google Keyword Planner is a common source.
What it is not
A count of AI prompts. Engines do not publish how often a prompt is asked, so search volume is a proxy for demand and not a prompt count.
SEO
Search engine optimization: the practice of improving a site so it earns visibility in the organic results of search engines.
Why it matters
Why it matters
It remains the base layer. Google states that SEO best practices stay relevant for its AI features, and a page that cannot be crawled or indexed has little chance of being cited.
What it is not
Paid search, and not the same as GEO. SEO is judged on rankings and clicks, while GEO is judged on whether a brand appears inside generated answers.
SERP
Search engine results page: the page a search engine returns for a query.
Why it matters
Why it matters
The SERP now often combines ordinary links with an AI summary, a question list, and other features, so a result's position is only part of how visible it is.
What it is not
A ranking list alone. Features such as AI summaries and snippets can sit above every ordinary result.
Unbranded query
A search or prompt that describes a need or category without naming any brand.
Why it matters
Why it matters
This is where discovery happens. When a buyer asks for the best option in a category, the engine decides which brands appear, and the brand has no say in the wording.
What it is not
A low-value query. It often has no brand in it precisely because the buyer is still choosing.
Zero-click search
A search that ends on the results page without a visit to any website.
Why it matters
Why it matters
AI answers resolve many questions in place, so a brand can influence a buyer without receiving a visit. That makes presence in the answer worth measuring on its own.
What it is not
A sign of a failed ranking. The reader got an answer, and what matters is whether the brand was part of it.
How AI search works
AI citation
A source that an AI answer links or lists as support for its text.
Why it matters
Why it matters
A citation is where an AI answer can send a visit and where a brand's own pages can act as evidence.
What it is not
A mention. An answer can mention a brand without citing its pages, and cite its pages without mentioning it.
AI mention
The appearance of a brand in the text of an AI answer, with or without a link.
Why it matters
Why it matters
A mention puts the brand in front of the buyer at the moment of the question, which is why mention rate sits at the center of AI visibility.
What it is not
A citation. A mention is about what the answer says, and a citation is about which sources it points to.
Brand hallucination
A false statement an AI engine makes about a specific brand, such as a wrong feature, price, or founder.
Why it matters
Why it matters
Buyers read these statements as fact. Spotting them takes tracking what engines say, and fixing them usually means publishing clear, consistent, verifiable facts that engines can retrieve.
What it is not
An omission. A brand that is missing from an answer has a visibility gap, and a brand described wrongly has an accuracy gap.
Chunking
Splitting a page into passages so that each can be retrieved on its own.
Why it matters
Why it matters
A passage that is pulled out of its page has to make sense alone. Sections that state their subject and answer clearly survive chunking better than ones that rely on earlier paragraphs.
What it is not
Something a site controls directly. Each engine chooses its own chunk sizes, so the aim is passages that stand alone and not a fixed length.
Cited source
A page or domain that an AI answer cites as evidence.
Why it matters
Why it matters
Looking at which domains engines cite for a category shows where their evidence comes from, including independent sites a brand does not own.
What it is not
Always the brand's own site. Engines often cite review sites, publications, and community pages instead.
Context window
The amount of text a model can consider at once, covering the prompt, retrieved sources, and its own reply.
Why it matters
Why it matters
It limits how much retrieved material can inform an answer, so only a small share of what is found reaches the model.
What it is not
Memory across conversations. The window covers the current exchange only.
Embedding
A list of numbers that represents the meaning of a piece of text so that similar meanings sit close together.
Why it matters
Why it matters
Embeddings let a system match a question to a passage by meaning, even when the two share no words.
What it is not
A readable summary. The numbers have no meaning to a person, only to the system comparing them.
Grounding
Tying an AI answer to retrieved sources so its claims can be checked against them.
Why it matters
Why it matters
Grounded answers cite sources, which gives a brand a place to appear as evidence and a reader a way to verify.
What it is not
Proof of accuracy. A grounded answer can still misread or misstate its source.
Hallucination
A fluent AI statement that is false or unsupported by any source.
Why it matters
Why it matters
The text reads as confident whether it is right or wrong, so a reader cannot tell by tone. Grounding reduces the problem without removing it.
What it is not
A deliberate lie or a rare glitch. It is a normal failure of systems that generate likely text.
Knowledge cutoff
The date after which a model has no training data.
Why it matters
Why it matters
Anything after the cutoff reaches an answer only through retrieval, so a model without search can describe a brand as it was, not as it is.
What it is not
The date the model was released. Release usually comes months after the cutoff.
Large language model
A neural network trained on very large amounts of text to predict and generate language.
Why it matters
Why it matters
It is the engine behind most AI answers. What it knows without searching comes from training, and what it knows after searching comes from retrieved pages.
What it is not
A database of facts. A model generates likely text, which is why it can state something false with confidence.
Non-determinism
The property that the same prompt can return a different answer each time it is run.
Why it matters
Why it matters
One answer is a single draw, not a result. Reliable measurement runs each prompt many times and reports the pattern across answers.
What it is not
Randomness with no pattern. Across enough runs, the same brands tend to appear at steady rates.
Position bias
The tendency of a model to give more weight to items at the start or end of a list or context than to those in the middle.
Why it matters
Why it matters
It is one reason the order of items in a source, or in an answer, can matter beyond their content.
What it is not
A fixed effect. Its strength differs between models and between kinds of input.
Prompt
The question or instruction given to an AI model.
Why it matters
Why it matters
Prompts are the unit of AI search, in the way keywords are the unit of classic search. Small changes in wording can change which brands an answer includes.
What it is not
A keyword. Prompts are usually full sentences and carry context, such as a role, a budget, or a location.
Query fan-out
A technique in which an AI engine splits one question into several narrower searches, runs them, and combines the results into one answer.
Why it matters
Why it matters
A brand can be absent from the question a buyer typed and still appear because it matches one of the narrower searches behind the scenes. Google describes both AI Overviews and AI Mode as able to use the technique, and it explains why a page that covers a topic from several angles has more chances to be retrieved.
What it is not
A ranking factor a site can switch on. The sub-queries are generated by the engine and are often not shown to the reader.
Recency bias
The tendency of retrieval-based answers to favor pages that are recent or recently updated.
Why it matters
Why it matters
For questions where recent information matters, a fresh page can be favored over an older and otherwise stronger one.
What it is not
A rule that new always wins. How much recency counts varies by engine and by the kind of question.
Reranking
A second pass that reorders retrieved passages by relevance before a model uses them.
Why it matters
Why it matters
Retrieval can return many candidates, and reranking decides which few reach the model. A passage that is retrieved but ranked low may never shape the answer.
What it is not
A visible ranking. The reordering happens inside the engine and is not shown to the reader.
Retrieval-augmented generation
A method in which a model retrieves relevant documents first and then writes its answer from them.
Why it matters
Why it matters
It is the reason a page published today can appear in an answer today. Retrieval pulls current pages into the answer, so what a site publishes and how it can be read both count.
What it is not
The model's memory. Without retrieval, an answer draws only on training data.
Semantic search
Search that matches the meaning of a query instead of only its exact words.
Why it matters
Why it matters
A page can be retrieved for a question it never repeats word for word, so clear coverage of a topic counts for more than repeating a phrase.
What it is not
Permission to ignore wording. Clear, specific language still helps a system judge what a passage is about.
Token
The unit of text a model reads and writes, usually a word or part of a word.
Why it matters
Why it matters
Models measure input and output in tokens, which sets how much text fits in the context window.
What it is not
A word. Common words are often one token, and long or rare words split into several.
Training data
The text a model learned from before it was released.
Why it matters
Why it matters
It shapes what a model recalls about a brand without searching. A brand that was widely discussed in the sources used for training is more likely to be recalled unprompted.
What it is not
Something a brand can update today. Publishing a page changes what retrieval can find, not what a trained model already holds.
Vector search
Retrieval that finds the passages whose embeddings sit closest to the embedding of the query.
Why it matters
Why it matters
It is one way retrieval systems find candidate sources for an answer, matching on meaning instead of exact words.
What it is not
The only retrieval method. Many systems combine it with keyword matching.
Measurement
AI referral traffic
Website visits that arrive from a link inside an AI answer.
Why it matters
Why it matters
It is the part of AI influence that shows up in web analytics. It is a different layer from visibility, which asks whether the brand appeared in the answer at all.
What it is not
A measure of AI visibility. Many buyers read an answer, never click, and act later, so traffic undercounts influence.
Answer volatility
How much the answers to the same prompt change from run to run or from day to day.
Why it matters
Why it matters
High volatility means a single reading is unreliable. It also means a small change in a rate may be noise and not progress.
What it is not
A flaw in the measurement. It is a property of the engines, and good measurement reports it.
Average position
Where a brand sits, on average, in the ordered list of brands an AI answer gives.
Why it matters
What it shows
Whether a brand tends to come first, in the middle, or last when an answer lists several options. A lower number means an earlier place in the list.
What it does not show
Anything about answers that do not mention the brand. Average position is calculated only across answers where the brand appears, so it is read next to mention rate.
Baseline
The first measurement of how a brand appears in AI answers, taken before any change so later results have something to compare against.
Why it matters
Why it matters
Without a baseline, an improvement cannot be told apart from drift. It also shows which engines and topics need the most work.
What it is not
A target. It records the starting point, and the same prompts and engines must be reused for later readings to compare.
Citation rate
The share of AI answers that cite at least one page from a brand's own domains.
Why it matters
What it shows
How reliably a brand's own pages show up as evidence. The count is answers, not sources: an answer counts once if it cites the brand's domains, however many other sources it lists. A brand can be mentioned often and still be cited rarely.
What it does not show
How large the brand's slice of the citation pool is. That is citation share. It also does not show whether the cited page was the best source for the question, or whether a visitor followed the link.
Citation share
The share of all citations in a set of AI answers that point to a brand's own domains.
Why it matters
What it shows
How much of the citation pool a brand holds. Every source cited across the selected answers counts once, and the brand's share is the portion that links to its own domains. Competitors and third-party sites fill the rest.
What it does not show
How many answers cite the brand at all. An answer that cites ten sources and one of them is the brand's adds little to citation share but still counts fully toward citation rate. It also does not show whether the cited page was the best source for the question, or whether a visitor followed the link.
Competitive benchmarking
Measuring a brand and its competitors on the same prompts and engines so that the results are comparable.
Why it matters
Why it matters
A rate means little alone. Seeing where competitors are mentioned and a brand is not shows which prompts and sources to work on.
What it is not
A comparison across different prompt sets. Results are only comparable when every brand is measured the same way.
Engine coverage
The set of AI engines included in a measurement.
Why it matters
Why it matters
Engines draw on different sources and behave differently, so a result from one engine does not carry over to another. Coverage decides how much of the market a figure describes.
What it is not
A count to maximize. The right engines are the ones the audience uses.
Mention rate
The share of tracked AI answers in which a brand is mentioned.
Why it matters
What it shows
Whether a brand appears at all. If a prompt set produces 100 answers and the brand is mentioned in 22 of them, the mention rate is 22 in 100.
What it does not show
How the brand is described, where it sits in the answer, or which sources the engine cited. Mention rate is read together with the other three measures.
Prompt set
The fixed list of buyer questions a brand tracks across AI engines, run repeatedly so results can be compared over time.
Why it matters
What it shows
What is being measured. Every rate in a report is only as meaningful as the prompts behind it, so a prompt set covers the questions a real buyer asks about the category, not only the brand's own name.
What it does not show
Independent observations. Repeated runs of one prompt are repetitions of the same question, not separate buyers.
Prompt tracking
Running a fixed set of prompts on a schedule and recording how a brand appears in the answers.
Why it matters
Why it matters
A fixed prompt set makes results comparable over time and across brands, so a change in the numbers reflects a change in the answers and not a change in the questions.
What it is not
A one-off test. Because answers vary run to run, a single pass says little, and the value comes from repeated runs on the same prompts.
Share of voice
A brand's mentions as a share of all brand mentions across the same set of tracked answers.
Why it matters
What it shows
How much of the conversation a brand holds against the others in its category. If the tracked answers contain 200 brand mentions and one brand accounts for 50, that brand's share of voice is 25%.
What it does not show
Whether the mentions are favorable or whether any answer cited the brand's own pages. The comparison is only meaningful when every brand is measured on the same prompts and engines.
Tracked answer
One recorded answer from one engine to one prompt on one run.
Why it matters
Why it matters
It is the unit that rates are counted over. Mention rate, citation rate, and share of voice all divide by answers, so the answer count sets how much weight a figure can carry.
What it is not
An independent observation. Repeated answers to the same prompt are repetitions, so a large answer count from few prompts is less precise than it looks.
Buyer behavior
Buyer journey
The steps a buyer takes from first noticing a problem to making a purchase decision.
Why it matters
Why it matters
AI answers now appear at several steps, from the first question about a problem to the final comparison between vendors.
What it is not
A straight line. Buyers loop back, ask follow-up questions, and compare in several places at once.
Category entry point
A situation or need that leads a buyer to start thinking about a category.
Why it matters
Why it matters
Entry points are a good source of the unbranded prompts worth tracking, because each one is a moment when a buyer asks an engine for options.
What it is not
A keyword. An entry point is a situation, such as a deadline or a change in the team, and it can produce many different queries.
Dark funnel
The part of a buying journey that happens where a company's analytics cannot see it, such as private chats, peer advice, review sites, and AI assistants.
Why it matters
Why it matters
A buyer who asks an AI assistant for a vendor shortlist leaves no trace in marketing attribution. By the time that buyer first contacts a company, part of the decision has already been made. Measuring what AI engines say about a category is one of the few ways to see inside that stage.
What it is not
The same as dark social, which is only the private-sharing part of it. It is also not unknowable: tracking AI answers makes one slice of it visible.
Dark social
Content sharing through private channels such as messaging apps and email, where the source of a visit goes untracked.
Why it matters
Why it matters
A link pasted into a chat arrives as direct traffic, so the share that led to it is invisible to analytics.
What it is not
The whole dark funnel. It also leaves out offline talk, review sites, and AI assistant queries.
Shortlist
The small set of vendors a buyer decides to evaluate seriously.
Why it matters
Why it matters
Many purchases are decided inside the shortlist, so appearing in the AI answer that forms it is worth more than ranking well after it is set.
What it is not
A list the buyer announces. It forms privately, often before any contact with a vendor.
Technical access
AI crawler
An automated program that fetches web pages for an AI company, to train models, build a search index, or answer a live request.
Why it matters
Why it matters
Different bots serve different purposes, so allowing one and blocking another changes what a model learns versus what an assistant can retrieve live.
What it is not
A single bot. Each company runs several, and each is controlled by its own name in the site's access rules.
Crawlability
How easily automated crawlers can reach and fetch a site's pages.
Why it matters
Why it matters
A page that cannot be fetched cannot be cited. Blocked paths, broken links, and pages with no inbound links all reduce crawlability.
What it is not
Indexing. A crawler can fetch a page that is never stored in the index.
Indexability
Whether a page that has been fetched is allowed and able to be stored in a search index.
Why it matters
Why it matters
For Google's AI features, a page must be indexed and eligible to appear in Search with a snippet before it can be a supporting link.
What it is not
Crawlability. A page can be reachable and still be excluded from the index by a tag or a quality decision.
JavaScript rendering
Building a page's content in the browser with JavaScript instead of sending it ready-made from the server.
Why it matters
Why it matters
Some crawlers do not run scripts, so content that appears only after rendering can be invisible to them. Serving key text in the initial HTML avoids the risk.
What it is not
A problem in itself. Many sites render well, and the point is to check what a crawler receives.
llms.txt
A proposed convention for a plain-text file that points language models to a site's most useful pages.
Why it matters
Why it matters
It gives a site a way to offer a clean list of key pages and summaries to tools that choose to read it.
What it is not
A formal standard, or a replacement for robots.txt and a sitemap. It does not control access.
robots.txt
A file at the root of a site that tells crawlers which paths they may fetch.
Why it matters
Why it matters
It is the first place to check when a brand is missing from AI answers, because a rule that blocks a crawler can keep a page from being read at all.
What it is not
A security control. It is a request that well-behaved crawlers follow, and it does not hide a page from anyone.
Sitemap
A file that lists a site's pages to help crawlers find them.
Why it matters
Why it matters
It helps new and deep pages get discovered, and it signals when pages were last changed.
What it is not
A guarantee of crawling or indexing. It is a hint and not an instruction.
Agents and protocols
Agentic search
Search in which an AI agent plans and runs several searches and reads sources across steps before giving an answer.
Why it matters
Why it matters
It goes further than a single retrieval round: the agent reads, decides what is missing, and searches again, so a brand has more chances to be found and more places to be dropped.
What it is not
The same as query fan-out. Fan-out runs sub-queries together, and agentic search adapts its next step to what it has read.
AI agent
Software that uses an AI model to pursue a goal through several steps, calling tools along the way.
Why it matters
Why it matters
Agents research, compare, and act for a user, so a brand may be evaluated by software before any person sees it.
What it is not
A chatbot that answers one question. An agent plans, uses tools, and continues until the task is done.
Model Context Protocol
An open standard that lets AI applications connect to external tools and data through a common interface.
Why it matters
Why it matters
It lets an assistant read live data from a service instead of relying on a copy, which opens a route for brands to expose accurate information directly.
What it is not
A search engine or an index. It is a way to connect tools, and what an agent does with them is up to the agent.