Glossary
The vocabulary of AI visibility.
Every entry opens with the definition, then explains the mechanism behind it. Where a claim needs evidence, it links the spec or the measurement it came from.
#
A
A/B Test
A controlled comparison of two versions, and why an assistant's answer has to be measured before and after instead of split.
Agentic Search
An assistant that runs its own sequence of searches and decides what to do next, and what that changes for a page that wants to be read.
AI assistant
The program that writes one answer from a model's memory and a few fetched pages, and why the fetched pages are where a business gets in.
AI Citation
The link an assistant attaches to a sentence it took from your page, decided passage by passage, and distinct from being recommended.
AI crawler
The fetchers behind training, search and live answers, told apart by their tokens, and what one of them was handed by 9,483 sites.
AI Ranking
Where you fall in a list the assistant writes anew each time, and why that has to be measured across a panel rather than read once.
AI recommendation
When the assistant names you as the choice rather than the source, what it had to establish first, and how often that happened in a 9,240-business probe.
AI Search
Search that returns a written answer instead of a list, the indexer behind each product, and the one robots.txt confusion that costs sites their place.
AI Visibility
How often AI assistants name a business when asked the questions its customers ask, measured as a score from 0 to 100.
Alt Text
The words a page offers in place of each picture, written to the HTML standard and read by every client that never loads the image.
Anchor Text
The words a link is made of, read as a label for where it goes, and what is and is not known about their effect on an assistant.
Answer Engine
A system that replies to a question with a composed answer rather than a list of links, and the one structural thing that follows from it.
Answer Engine Optimization (AEO)
The older name for the same work GEO describes: making the answer, not only the question, present on the page where a machine can lift it.
Answer Page
One question, answered in full at the top of its own page, and the audit figure showing how rarely small-business sites are built that way.
Anthropic
The company behind Claude, the assistant this site probed 9,240 businesses against, and why its numbers stay labelled as Claude's.
Article Schema
Structured data that says a page is a piece of writing with a headline, an author and dates, and why that is a different job from describing the business.
Artificial Intelligence (AI)
The outermost word for systems that learn or reason, narrowed to the one kind that matters here: an assistant that fetches pages and names businesses.
Author Profile
The page that turns a byline into a person: identity, qualification and links to the same name elsewhere, declared once and linked from every article.
B
Backlink
A link from another site to yours, what the three rel values do to it, and the boundary between earning a fetch and earning a citation.
Baseline Measurement
The first reading of the question panel, taken with the questions, assistants and scoring frozen so that the second reading means something.
Benchmark
A reference value from outside the thing measured, in its marketing sense and its research sense, and the conditions under which one is fair.
Bounce Rate
The share of visits that end on the landing page, why the number depends on the tool, and the whole class of AI answers it never sees.
Brand Awareness
Whether buyers know a business exists, and the newer question beside it: whether the assistant they ask knows, which moves on its own.
Brand Mention
The bare appearance of a business name in text it did not write, linked or not, and the version of it that lives inside an assistant’s answer.
Brand Monitoring
Tracking where a business is named, and the channel that broke the old tooling: answers that are generated fresh and leave no page to scrape.
Breadcrumb Navigation
The row of links that places a page in the site’s hierarchy, in a form a person reads and a form a machine reads, and what each learns from it.
C
Call-to-Action
The sentence that asks the reader to act, what makes it work, and why it is the one part of a page an assistant never lifts into an answer.
Canonical Tag
The head element that names one preferred address for duplicated content, what the RFC lets a search engine do with it, and how far it is obeyed.
Case Study
One piece of work, told with the before figure, the after figure and the method, and why that is the content an assistant can quote from a services site.
ChatGPT
OpenAI’s assistant, the three crawler tokens behind it, and why blocking the training bot does not change what ChatGPT search can cite.
Citability
The property of a passage that lets an assistant lift it as the answer and attribute it: self-contained, specific, reachable, and clearly someone's.
Citation Rate
The share of answers that cite you across a fixed set of questions, and why it is usually lower than the share that merely mention you.
Claude
Anthropic’s model and assistant, and the only engine this site has probed at scale: 14.9% of 9,240 businesses named on one question each.
Click-Through Rate (CTR)
Clicks divided by impressions, and why a recommendation inside an AI answer can happen with no click to count.
Comparison Article
Options side by side on named criteria. The table row is what gets quoted; the stated method is what gets it trusted.
Competitive Analysis
Who an assistant names instead of you, with the reasons it gives. In one probe, 99.4% of those names were businesses, not directories.
Content Marketing
Publishing useful material so others point at you. Assistants point at sentences, and most small-business pages have none shaped for that.
Content Strategy
Which questions, which pages, what evidence, how measured. Half of audited sites chose the question and never placed the answer.
Context Window
Everything a model can read while it answers, including the pages it just fetched. Your page is competing for a budget, and only compact passages survive.
Conversational Search
Ask a question, get a written answer that names a few options. The crawler is the same; the value of a passage that answers alone went up.
Conversion Rate
The share of visitors who do the thing the page is for. Measured on your site, after the AI recommendation this site measures has already happened.
Core Web Vitals
LCP, INP and CLS describe a browser session. An AI crawler has no screen and often no script engine, so a page can score well and serve nothing.
Crawling
Automated fetching of pages. Robots.txt governs the indexers, not the on-demand fetch; and 7.2% of audited sites served a bot wall instead of a page.
Customer Journey
The steps from first need to purchase and referral, each with its own questions, and each now sometimes put to an assistant instead of a search box.
D
Digital PR
Earning coverage on other sites; the link speaks to a search engine, the sentence around it is what an assistant reads.
Domain Authority
A vendor's 0 to 100 estimate of a whole site's ranking strength, built from links; not a figure any search engine or assistant publishes.
Duplicate Content
One page reachable at several URLs; the crawler picks a copy unless you state one, and links and citations split across the rest.
Dwell Time
How long a searcher stays before going back to the results; visible only to the search engine, and absent when an assistant answers without a click.
E
E-E-A-T
Google's four-part frame for judging a page: experience, expertise, authority and trust, with trust carrying the rest; a rater's checklist, not a score.
Entity
One identifiable thing, separate from the words that name it; a machine resolves three spellings to one business, or fails to.
Entity Optimization
Making one business unambiguous to machines: name, address, phone, schema and profiles all saying the same thing, on every page they appear.
Event Schema (Event)
Structured data that gives an event a name, a start time and a place a machine can read, one event per page.
Evergreen Content
A page whose answer stays correct for years because the question does not change; specific, maintained occasionally, never faked fresh.
F
Fact-Checking
Verifying a claim against evidence; on the web both an editorial habit and a markup type, and something an assistant does only against what a site serves.
FAQ Page
Customers' real questions with the answer written directly under each; the structure an assistant lifts from, and one most small-business sites lack.
FAQ Schema
JSON-LD that repeats a page's visible question-and-answer pairs for machines; the rich result it once earned is gone, the description remains.
Featured Snippet
The passage Google lifts to the top of a results page, chosen by its systems from a page that answers the query in one place.
Fine-Tuning
Continued training of a model on a small example set to shape how it answers; it changes weights, not what the model can look up.
Freshness
How recently a page changed, as read from the header, the schema dates, the sitemap and the text; four signals that should say the same thing.
G
Gemini
Google's generative models and the assistant built on them; a site reaches it through training or grounding, both governed by one robots token.
Generative AI
Software that composes new text or media from a prompt rather than retrieving it; assistants pair it with retrieval so live pages feed the output.
Generative Engine Optimization (GEO)
The work of making a business more likely to be named, cited or recommended in the answers ChatGPT, Claude, Gemini and Perplexity write.
Generative Search
A search that answers in a written paragraph built from retrieved pages, with the ranked list moved behind the citations.
Glossary
A set of terms with one definition each, one page per term, written so the first sentence answers the question on its own.
Google AI Overviews
Google's generated summaries above the results, drawn from indexed pages with links back; no special markup, the usual Search controls.
GPTBot
OpenAI's training crawler token. Blocking it withholds pages from training; ChatGPT search and user fetches run under two other names.
Grok
xAI's assistant, inside X and as an app. This site has not measured it, so the entry states what carries over from every retrieval assistant and no more.
Guide Article
A long-form page that walks one task from start to finish, built so each step is a heading with its own liftable answer.
H
Hallucination
A confident statement with nothing behind it. For a business, the gap it fills is often one the business's own pages left open.
Heading Hierarchy
The nesting of h1 to h6 that lets a page's headings read as an outline, and that tells a machine which passage answers which question.
HowTo schema
Step-by-step markup from schema.org: what the properties mean, why Google dropped the rich result, and how to tell a HowTo from an FAQ.
hreflang
The annotation that tells a search engine which language a page is in and where its siblings live, and the return-link rule that most often breaks it.
HTTP Status Code
The three digits at the top of every response, what each class tells a crawler, and why the code alone does not say whether the site came through.
HTTPS encryption
HTTP over TLS: what the connection actually proves, how lightly Google weighs it, and why answering on both schemes hands the engine a choice you should have made.
I
Indexing
The stage between the fetch and the search result, why it is never guaranteed, and the two roads by which a page reaches an assistant's answer.
Internal linking
Links between your own pages: what a crawler will follow, why an unlinked contact page does not exist to it, and where nofollow does not belong.
J
K
Keyword
The phrase a person types, and what happens to it when the search box becomes an assistant and the phrase becomes a question with a place in it.
Keyword cannibalization
Two of your pages written for one query: how the engine chooses between them, how to tell overlap from coverage, and what the mechanism does in an AI answer.
Keyword research
Finding the words buyers actually use, why there is no search-volume figure for a question put to an assistant, and what to measure instead.
Knowledge Base
Support articles and retrieval corpora share one shape: findable from the question, complete once found. Your website is the corpus for the answer written about you.
Knowledge Cutoff
The date a model’s training data stops, why it is the wrong thing to blame for being unknown, and what actually decides it.
Knowledge Graph
A database of things rather than pages, how an engine decides two records are the same business, and what inconsistent contact details do to that decision.
Knowledge Panel
The box beside the results when the engine knows exactly what you mean, where it is filled from, and the few-names answer that is its counterpart in an assistant.
L
Landing Page
A page built for one arriving audience and one action, and what happens when the reader is an assistant looking for a fact rather than a form.
Language Model
A probability model over text that writes by predicting the next token, and why the pages it fetches matter more to you than the text it was trained on.
Large Language Model (LLM)
The neural network under ChatGPT, Claude and Gemini, and the two separate paths, training and retrieval, by which your business reaches its answers.
Link Building
Earning references from other sites, and the point in an assistant's process where a link stops mattering and the passage on the page takes over.
llms.txt
A proposed reading list for language models at /llms.txt, set beside the robots.txt standard it is often mistaken for, with the audit's adoption figure.
Loading Time
Seconds from request to a usable page, weighed as Google weighs it, and the separate question of what a script-free crawler receives in that time.
Local SEO
Showing up for a service in a place: three homes for one identity, the audit's consistency figures, and what an assistant adds to the requirement.
LocalBusiness Schema
The schema.org type for a business at an address, what Google asks of it, and the audit's finding that most existing markup describes the website instead.
Long-Tail Keyword
The specific, low-volume query that says exactly what someone wants, and the reason questions to an assistant almost always take that shape.
M
Machine Learning
Programs that improve against examples instead of following written rules, the three ways they are trained, and where language models fit among them.
Mention Rate
The share of answers that name you at all, why two honest denominators give two numbers, and the probe that shows the gap between them.
Meta Description
The head-of-page summary Google may or may not show, what its own documentation says about that choice, and the one reader who never sees it.
Meta Llama
Meta's openly weighted model family, and why a model anyone can deploy leaves a business one lever that reaches every deployment: the page itself.
Meta Tag
A name-and-value element in the page head; most are ignored, and the robots family now decides what AI Overviews may excerpt.
Microdata
Structured data written as attributes on the visible HTML: five attributes, one syntax of three, and a trade-off against JSON-LD.
Microsoft Copilot
Microsoft's assistant, grounded in Bing and present in Windows, Edge and Office; this site has not probed it and says so.
Mistral
A French model maker whose models mostly run inside other people's products; the deployer, not the vendor, decides what gets fetched.
Mobile Optimization
Making the page work on a phone, which is the version Google indexes; the same failures blind an AI crawler.
Multimodal AI
A model that takes images and text together; useful to the customer holding a phone, and no help for facts your site keeps only in pictures.
N
Natural Language Processing (NLP)
The field of making machines read and write language; two of its tasks, entity recognition and question answering, decide what an assistant takes from your site.
Neural Network
Layers of weighted units trained to map input to output; it stores no facts about you, which is why assistants go and fetch your page.
O
Off-Page Optimization
Links, mentions, listings and reviews elsewhere on the web; an assistant uses them to corroborate a business, not to count votes.
On-Page Optimization
Everything inside the page a machine reads; the audit found the words present on most sites and the answerable form absent.
Open Graph
The og: meta tags that shape a link's share card; not a ranking signal, but one more account of the page that should agree with the rest.
OpenAI
The maker of ChatGPT, met by a website as three crawlers with three jobs; blocking the training one leaves the search one untouched.
Organic Traffic
Visits from unpaid search results, sorted by referrer; being named in an AI answer produces direct visits, referrals or nothing, never an organic one.
Organization Schema
The markup that says who publishes a site: name, logo, address, profiles. Most small-business sites carry schema that describes the website instead.
P
Page Title (Title Tag)
The one line a tab, a bookmark, a search result and an assistant all read before anything else on the page.
Perplexity
An answer engine with two fetchers: one builds an index and obeys robots.txt, one fetches for a user and generally does not.
Pillar Content
The long parent page in a topic cluster. It is cited section by section, so its length matters less than whether each section answers on its own.
Position Zero
The answer box above result one. Nothing marks a page up for it; a passage that answers in one place earns it, and the same passage is what assistants lift.
Product Schema
Markup for one item with a price and a rating. Services are not products, and a price in a PDF is a price no fetcher reads.
Prompt
Everything a model works from, including the pages it fetched a moment ago. A web page is part of someone else's prompt.
Prompt Engineering
Changing the words, not the model, against a test set. It works on prompts you write; a customer's prompt is answered from your pages.
Q
R
Ranking Factor
A signal an engine orders results by. Google names systems, not weights, and a factor only decides whether an assistant sees your page, not whether it names you.
Reach
Distinct people who saw a message. No assistant reports it, so for AI answers the measurable stand-in is how often you are named.
Responsive Design
One document, one URL, any screen. Invisible to a crawler with no screen, which is why JavaScript-only rendering is the failure that matters there.
Retrieval-Augmented Generation (RAG)
The model reads pages fetched for this question and cites spans from them. Your page is re-read every time, so what it serves now is what gets quoted.
Review Schema
A rating in a form a machine can parse. Google shows no stars for a business's own reviews, yet most sites give a fetcher no rating signal at all.
Rich Snippet
A search result with extra detail read from your markup; eligibility, never a guarantee, and the same data an assistant uses to identify you.
robots.txt
The file that asks crawlers to stay out of paths; how the longest match wins, why it is not a lock, and which AI tokens a rule must name.
S
Schema.org
The shared vocabulary machines read a page with; the difference between the words and the format, and the type a business should actually use.
Search Engine Crawler
The program that fetches pages for an index, and the ways an assistant's answer-time fetcher differs from Googlebot.
Search Engine Optimization (SEO)
Making a site easy to crawl, understand and rank; the three layers it works in, and the point where a rank stops being the thing measured.
Search Engine Results Page (SERP)
The ranked page an engine returns and everything around the list; why a position on it is not the same thing as being named in an answer.
Search Intent
What the searcher is trying to do; the usual four kinds, and how a spelled-out question makes intent a test the page passes or fails.
Search Volume
A modelled monthly count for a string of words; useful for choosing pages, silent on the questions people ask an assistant.
SearchGPT
OpenAI's search prototype, now ChatGPT search; the crawler that feeds it and the one robots.txt rule that decides whether you appear.
Semantic Search
Retrieval by meaning: how vectors stand in for words, and why a self-contained specific sentence is what gets found.
Sentiment Analysis
Reading the tone of the sentence an assistant names you in, because a mention, a caveat and a recommendation are not the same result.
Share of Model
The proportion of an assistant’s answers, across a fixed question set, in which a business is named, and why it is measured per assistant.
Share of Voice
Your slice of every business name an assistant gives for a set of questions; relative where the score is absolute.
Source Citation
The link an assistant hangs on a claim; how the fetch produces it, and why a cited page and a named business are two different wins.
Structured Data
Machine-readable markup that names the business, address and hours on a page; most sites have some, and most of it describes the website instead.
System Prompt
The operator's standing instructions to a model: role, rules and format, applied to every question before the user has typed one.
T
Target Audience
The people a business means to reach, defined by need and place; in visibility measurement, the people whose questions the panel is built from.
Technical SEO
The work of making sure a crawler can reach, read and index a site; the same gate an AI assistant's fetcher has to pass before content matters.
Temperature (AI model)
A sampling setting that decides how often a model picks a less likely next token; the reason one AI answer is a sample and a panel is a measurement.
Token
The fragment of text a model actually reads; pages, prompts and answers are all counted in them, which makes a long page a budget rather than a virtue.
Topic Cluster
A pillar page and its question pages, linked both ways so each question has one home; the shape that leaves an assistant a passage to quote.
Topical Authority
Being treated as a dependable source on a subject through coverage rather than links, and the failure mode of chasing it with volume.
Training Data
The corpus a model was fitted to before release, fixed at a cutoff; a separate channel from the live retrieval that decides which businesses get named.
Transformer Model
The architecture under every current large language model, which relates each token to every other; it rewards a claim that sits next to the name it belongs to.
U
Unique Selling Proposition (USP)
The one checkable reason to pick a business over its rivals, stated so an assistant can lift it into the clause that follows your name.
URL Structure
How a site's addresses are organised; legible paths help a reader, but one page at one stable address is what keeps a citation alive.
V
Vector Database
The store that holds embeddings and returns the nearest ones to a question; it works per chunk, so a fact far from the name it belongs to arrives as an orphan.
Vector Embedding (Embedding)
Numbers a model assigns to a text so that similar meanings sit close together; the reason a specific sentence is found and a page of praise is not.
Visibility Score
The 0 to 100 number this site gives a business: the share of a fixed panel of twelve buyer questions in which an AI assistant named it.
Voice Search
Asking a device aloud and hearing one answer back. Why the passage the engine picks decides everything, and what speakable markup covers.