The vocabulary behind the report
AI Search Glossary
24 terms that decide whether AI names your business
Every term below is one the AI engines act on, defined in plain English in one sentence, followed by why it matters and, wherever the record allows, a dated finding from AI Search Engine Optimization’s own measurement. Read it to hold any report, including ours, to a fixed meaning.
The definitions are the ones Rank Intelligence Engine uses when it measures a site across ChatGPT, Perplexity, Gemini, Copilot and Google’s AI Overviews and AI Mode, and the findings come from the weekly measurement of this site: 26 buyer and research questions on four AI platforms.
30 minutes with Phil. Clear next steps. No obligation.
24 termsDefined in one sentence each
Dated findingsFrom a weekly measurement of this site
Four groupsCounting, sources, verification, and the business
What the engines count
Six terms behind every number in an AI visibility report
Counts are only comparable when the thing counted is the same thing. These six are the units.
AI citation
An AI citation is a link or footnote inside an AI answer that names a specific page as the source of a statement. Citations are counted per page and per platform, separately from mentions, because an answer can quote a page without naming the business that published it.
Seen in the record: this site was cited eight times across 26 buyer and research questions over the seven days to September 13, 2026, all by ChatGPT, on two pages.
AI mention
An AI mention is the appearance of a business’s name in the text of an AI answer, with or without a link to its site. Mentions are what a buyer reads; citations are what the engine leaned on. The two move separately and are never reported as one number.
Seen in the record: named in two answers in the same window in which it was cited eight times.
Recommendation, or first call
A recommendation is an AI answer’s stated pick, the business it tells the buyer to contact first, as distinct from being one of several names in a list. Position inside the answer is recorded with every mention, because being fifth of five and being the first call are different results.
Seen in the record: on September 13, 2026 ChatGPT shortlisted five firms for a buyer wanting an agency that specializes in WordPress and named AI Search Engine Optimization as its first call.
Share of answers
Share of answers is the fraction of AI answers to a buyer question that name a given business, measured against every competitor the answers name, per platform and per date. The competitor set comes from the answers themselves, not from the business’s own list, because the businesses the AI recommends instead of you are the field to beat.
Seen in the record: the three most-named competing agencies on the recommendation questions appeared in 19, 14 and 12 answers in one seven-day window.
Buyer question, or prompt
A buyer question is the wording a prospect types into an AI assistant when looking for a provider or an answer, and it is the unit AI visibility is measured in, the way a keyword was for rankings. A tracked set pairs recommendation questions, where being named matters, with research questions, where being cited matters.
Seen in the record: 26 questions on four platforms, run weekly; a question’s answer on one platform is never read as its answer on all.
Answer position
Answer position is where in an AI answer a business is named, first, third or last in a list, averaged over every answer that named it. A low average with few mentions means the engine likes the business when it thinks of it and rarely thinks of it; the two figures are read together.
Seen in the record: an average position of 1.5 across the two answers that named this site in the seven days to September 13, 2026.
Where the answers come from
Six terms for the surfaces and the mechanics behind them
Each platform builds its answer differently, and the differences decide which page it cites.
Google AI Overview
Google AI Overview is the AI-written summary Google places above the organic results for many searches, with source cards drawn from pages in its index. It is a separate contest from the ten organic listings, and a page can rank on page one without appearing in it.
Seen in the record: on September 16, 2026 the AI Overview for the search this firm’s homepage targets cited three agency service pages.
Google AI Mode
Google AI Mode is Google’s conversational search tab, which answers in full prose from the same index and Knowledge Graph as the AI Overview but interprets the question on its own. The two features can read the same words differently, so a business is measured in each.
Seen in the record: on September 13, 2026 AI Mode answered a question about AI SEO services with software tools while the AI Overview and ChatGPT answered it with agencies.
ChatGPT Web Search
ChatGPT Web Search is the mode in which ChatGPT fetches live pages before answering and lists them as sources, as opposed to answering from what the model already knows. Only the search mode produces citations; the plain mode produces mentions from memory, and both are tracked.
Seen in the record: both modes built their agency shortlists from third-party lists, and both localized answers to a guessed city.
Grounding, or live retrieval
Grounding is the step in which an AI system retrieves pages from the web before it writes, so that the answer rests on documents rather than on memory alone. Engines that ground their answers can cite; a model queried without retrieval can only recall, which is why the two kinds of result are kept apart.
Where it is used: Google AI Overview and AI Mode, ChatGPT Web Search, Perplexity, Gemini and Copilot ground their answers; a model called through an interface with no search tool does not.
Source list
A source list is the set of pages an AI answer discloses as its evidence, with each page’s position in the list. It is the raw material for citation counts and for the placement plan, since the pages cited instead of yours are the ones to be on.
Seen in the record: 5 of 104 answers captured on September 13, 2026 returned no source list at all and were recorded as unmeasured rather than as zero.
Fan-out
Fan-out is the way an AI engine splits one question into several sub-questions, retrieves sources for each, and assembles the answer from the parts. Each section of an AI answer is therefore a sub-question a page could own, and a page that answers one cleanly is a page that gets cited.
Seen in the record: asked for the best AI SEO services, ChatGPT Web Search added a six-item section on what to look for that no one asked for; that section is a page waiting to be written.
What the engines have to verify
Six terms for the checks an engine runs before it trusts a site
AI engines name businesses they can resolve, read and quote. These are the six conditions.
Entity
An entity is a uniquely identifiable thing, a business, person, place or product, that an engine can tell apart from everything with a similar name. Search engines rank pages; AI engines recommend entities, and they name the ones they can resolve with confidence.
Seen in the record: when a business’s name is also its category term, the name is not detectable in answers, so its visibility is tracked by domain and two named identifiers instead.
Knowledge Graph record
A Knowledge Graph record is Google’s own entry for an entity: its identifier, types, description, image, official site and a confidence score, drawn on by AI Overview, AI Mode and Gemini when they decide whether a business exists and what it does. A weak or absent record is the most common reason a business that ranks is missing from Google’s AI answers.
Where it is measured: per business and per competitor, with the confidence score dated so movement can be seen.
Schema twin
A schema twin is the machine-readable copy of a fact that also appears in the visible text of the same page, written in structured data so an engine reads what a person reads. A figure asserted in schema with no visible twin is a claim the engine cannot check, and a visible claim with no twin is one it may not find.
Where it applies: every FAQ, every service, every claim in a definition block, and the organization record itself.
sameAs
sameAs is the structured-data property that lists the profiles and records that refer to the same entity as a page: the business’s social profiles, directory listings, and where they exist its Wikidata and Wikipedia entries. It is how an engine confirms that five references across the web are one business rather than five.
Where it applies: a closed set, identical on every page of the site, never a list that grows by accident.
Answer block
An answer block is a passage on a page that an AI engine can lift whole: it names its subject in the first sentence, makes a specific and verifiable claim, stands without the paragraphs around it, and is mirrored by the page’s structured data. Pages that carry one are the pages that get cited; pages that do not are summarized from someone else’s block.
Seen in the record: 101 blocks across eleven of this site’s own pages were scored on September 13, 2026; none was ready to lift whole and 73 were not extractable, and the rebuild went in the same day.
AI crawler access
AI crawler access is whether the crawlers behind ChatGPT, Perplexity, Gemini, Copilot and Google can actually fetch a site’s pages, tested by requesting the pages as each crawler rather than by reading the robots file. A firewall or hosting rule can block a crawler the robots rules allow, and the robots file will never show it.
Where it is checked: live, per crawler, before any content work, because nothing else in AI search happens until the engine can read the site.
Around the site, and back to the business
Six terms for what surrounds the site and what comes back from it
Most of what decides an AI recommendation lives off the site, and the only result that matters arrives back on it.
llms.txt
llms.txt is a plain-text file at the root of a site that lists its most important pages and what each is about, in a form written for AI systems rather than for browsers. It does not make a page citable on its own, but it tells an engine where to look first and what the site claims to be about.
Where it applies: kept in step with the sitemap and updated whenever pages are added or retired.
Third-party validation
Third-party validation is the evidence about a business that lives on sites it does not control: best-of lists, review platforms, directories, editorial coverage, community threads and video. AI engines assemble their recommendations largely from those sources, which is why no on-site change alone wins a recommendation question.
Seen in the record: across 26 questions on September 13, 2026, 16 of the 20 most-cited sources were third-party lists, comparison pages or directories, and none was an agency service page.
Brand detection
Brand detection is the rule a tracker uses to decide that an AI answer refers to a given business: its name, its aliases, its domain, or a pattern written for it. A wrong rule inflates a competitor’s count or hides your own, so the rule is set before anything is counted.
Seen in the record: two agencies whose names are the same three words in a different order are the kind of overlap a detection rule has to be written for.
Sentiment
Sentiment is a score for the tone an AI answer takes toward a business, from negative through neutral to positive, computed over every answer that names it. It only means something at volume; a score built on two mentions is noise and is reported as such.
Where it matters: a negative or inaccurate description in AI answers is a finding to fix on the sources the answer drew from, not on the answer.
Brand accuracy
Brand accuracy is whether an AI answer describes a business correctly: its services, its location, its audience and what it is known for. Inaccuracy usually traces to the sources the engine read, and it is invisible until someone reads the answers instead of counting them.
Seen in the record: the same buyer question was localized by ChatGPT to Charlotte, Philadelphia and Santa Fe on successive runs, with different local agencies added each time.
AI referral session
An AI referral session is a visit to a website that arrived from an AI assistant, identified by the referring domain in analytics, and it is the first business measure that follows a citation. It undercounts, because many AI answers are read and acted on without a click, which is why it is paired with a question asked at the point of inquiry.
Where it is completed: an inquiry form that asks every prospect which AI assistant sent them, by name, which this site’s forms do.
Where these terms are used
The same vocabulary,
on every page
These are the units in which an AI visibility audit reports a site, the measures the AI SEO services are judged by afterward, and the meanings every figure on this site is held to. When a proposal uses one of these words differently, ask which definition it means; when it cannot say, that is the finding.
The glossary is maintained beside the measurement it draws on, so a finding here is replaced when the next run changes it, and the date beside each one says which run it came from. The full set of analyses behind the measurement sits in Rank Intelligence Engine.
Your next move
See these figures
for your own site
Book a free 30-minute strategy call with Phil. Bring the site; leave knowing which of these 24 measurements you already have, which you have never seen, and which one to start with.