AI visibility glossary: every term defined in plain language
This glossary defines every term used on this site, in plain language, two or three sentences each. Definitions are written to stand alone, so any one of them can be quoted without the rest.
A to C
AEO (Answer Engine Optimization)
Work aimed at being the answer an AI or search engine gives, rather than a link it lists. One of several labels for the same discipline this site calls AI visibility.
AI crawler
A program an AI company runs to read web pages, either to collect training material or to fetch pages for live answers. Each has a name (GPTBot, PerplexityBot, ClaudeBot) that appears in server logs, and each can be allowed or blocked in robots.txt.
AI Mode
Google’s conversational search experience, which answers questions with generated text and follow-ups instead of a results page. It runs multiple related searches over Google’s index to build an answer.
AI Overviews
The AI-written summary Google shows above traditional results on many searches. It synthesizes web sources into a short answer, with links to some of them.
AI visibility
Whether AI assistants mention your business when customers ask them who to hire. The outcome all the acronyms in this glossary describe work toward.
Answer engine
Any system that responds to a question with a written answer instead of a list of links. ChatGPT, Perplexity, and Google’s AI surfaces all qualify.
Citation
An answer crediting the source it drew from, as a reference or link. A page about your business can be cited without your business being named, which is different from a mention.
Citation rate
The share of tracked prompts where an AI’s answer cited a source about your business, named or not. A supply-chain number: it shows whether the sources about you are being read.
Competitors
In tracking terms, the other businesses AI mentions on the same tracked prompts as yours. Watching them shows you which public sources currently win your market’s questions.
Core tracked prompts
The questions we already know customers ask in your industry, pre-built so tracking is useful from day one. The counterpart of custom prompts.
Crawl budget
A rough term for how much attention a crawler spends on a site before moving on. Mostly a concern for huge sites; local business sites rarely need to think about it.
Custom prompts
Questions specific to your business that you pick yourself for tracking: a niche service, a neighborhood, a question your phone calls always start with.
D to G
Entity
How machines model a real-world thing (a business, a person, a place) as distinct from any single web page about it. AI assistants evaluate your business as an entity, assembled from every source they can read about it.
GEO (Generative Engine Optimization)
Another label for AI visibility work, coined in a 2023 academic paper studying how content changes affect inclusion in generated answers. The term you’ll see most in research and tooling.
Generative engine
An academic term for systems that generate answers from retrieved sources, the way ChatGPT or Perplexity do. The “G” in GEO.
Google Business Profile (GBP)
Google’s listing for a local business: category, hours, photos, reviews, attributes. One of the most-read sources about any local business, by machines and humans alike.
GPTBot
OpenAI’s training crawler, identified by that name in server logs and robots.txt. Blocking it affects training data collection; OpenAI uses separate agents for live search fetching.
Grounding
Connecting a model’s answer to retrieved, current sources instead of memory alone. A grounded answer about your business is built from what’s readable about you today.
H to L
Hallucination
A model generating plausible details no source contains, like a service you don’t offer. It happens most where public information is thin, which makes it partly a signal problem you can influence.
Index
A search engine’s stored, organized copy of the web, used to find pages fast. AI assistants that search the live web do it through an index, their own or a partner’s.
Inference (in claims)
A conclusion reasoned from what’s known rather than observed directly. Honest AI visibility writing labels inferences as inferences; this site’s methodology requires it.
Knowledge cutoff
The date a model’s training data ends. Anything that changed about your business after the cutoff is absent from the model’s memory and reachable only through live lookup.
llms.txt
A proposed convention: a plain-text file summarizing a site for AI systems, like robots.txt for permissions. Adoption by platforms is limited and unevenly documented; treat claims about its power skeptically.
LLM (large language model)
The model type behind modern AI assistants, trained on large text collections to generate language. ChatGPT, Claude, and Gemini are products built on LLMs.
Local pack
The boxed map result with two to three businesses that Google shows for local searches. A different surface from AI answers, still heavily used, fed largely by Google Business Profiles.
M to P
Mention
Your business named in the text of an AI answer. The atomic unit of AI visibility: the thing tracking counts, and the thing this site’s name is about.
Mention rate
The share of tracked prompts where AI named your business in its answer. The headline tracking number, defined in depth alongside its two companions, verifiable by re-reading the stored answers behind it.
NAP consistency
Name, address, phone number agreeing everywhere they appear: site, profiles, directories. Disagreement makes a machine’s choice between sources a coin flip, and makes you riskier to recommend.
Prompt
The text typed to an AI assistant; in this site’s context, a customer’s question. Marketing copy on this site says “questions your customers ask” because that’s what local prompts are.
Prompt sampling
Asking a designed set of questions rather than one, so the result is a sample instead of an anecdote. Our scan uses ten prompts per business for exactly this reason.
Q to S
Quotable block
A two-to-three sentence passage written to make sense with zero surrounding context, so an AI can lift it verbatim as an answer. A structural habit this site both teaches and uses.
RAG (retrieval-augmented generation)
The technique of fetching relevant sources first and generating the answer from them. It’s how assistants answer about things too small or too recent to be in training data, including most local businesses.
Retrieval
The live-lookup half of how AI knows about your business: searching, reading current pages, building the answer from them. The half you can influence in weeks rather than model-release cycles.
Robots.txt
The file on a website that tells crawlers what they may read. Where AI crawlers get blocked, sometimes deliberately, often by a security default nobody remembers choosing.
Schema (structured data)
Machine-readable labels added to pages, marking up what a business is, where it is, what it offers. Helps machines parse your site; claims about its effect on AI answers specifically are mostly unproven.
Share of voice
Your business’s percentage of all business names given across a set of tracked prompts. Meaningful only when the prompt list is disclosed, because the set chosen defines the number.
T to Z
Tracked prompts
The questions customers ask AI assistants, tracked so you can see whether your business gets mentioned in the answers. The unit of measurement behind every number in an AI visibility dashboard.
Training data
The frozen collection of text a model learned from before release. Big, stable, widely referenced sources make it in; most local businesses barely do, which matters less than owners fear.
User agent
The name a program presents when requesting a web page, visible in server logs. How you tell which AI crawlers actually visit your site.
Zero-click search
A search that ends without the searcher clicking any result, because the page (or the AI summary on it) answered them. A growing pattern that makes being in the answer matter more than being in the links.
Terms we left out on purpose
A few phrases circulate in this industry that we won’t define because they don’t hold a definition: “AI whisperer,” “algorithm hacking,” “guaranteed placement,” and any acronym invented by a single vendor to name a service only they sell. If a term only ever appears in one company’s sales material, it’s a brand, not a concept, and this glossary sticks to concepts.