The AI Search Glossary: Terms Marketers Keep Mixing Up

Definitions and terms

AEO (Answer Engine Optimization) — Structuring content so a system that extracts one direct answer — a featured snippet, a voice result, an AI Overview, a chat answer — can pull yours out cleanly. Predates generative AI; see what AEO actually means.

AI search visibility — How often, how accurately and in what company AI engines name, describe and cite a brand when people ask about its category. ContextOwl's own working definition, not an industry standard. See the full explainer.

Citation — A source link or named reference attributed to a claim in an AI answer. A citation doesn't have to point to the brand's own site — it can credit a third party discussing that brand. See mentions vs citations vs recommendations.

Citation share — The percentage of citations for a given query (or "grounding query") that go to one site, out of all citations shown across every site for that same query. Microsoft added this as a named metric to Bing Webmaster Tools' AI Performance report on 16 June 2026.

Entity — A specific, identifiable thing a search or AI system can recognize and reason about distinctly from plain text — a company, a product, a person — typically connected to structured facts rather than just a string of words. Entity clarity (consistent naming, structured facts) is cited as a factor in how reliably engines describe a brand correctly, though the evidence for how much it moves any single metric is thin.

Fan-out (query fan-out) — The technique generative engines use of running several related searches for one user question, rather than one search, then assembling an answer from the combined results. Google's guide to its own AI features documents this directly for AI Overviews and AI Mode.

GEO (Generative Engine Optimization) — Work aimed at getting quoted or cited inside a generated, synthesized AI answer, as opposed to ranking a link. The term comes from a 2023 Princeton/Georgia Tech paper (accepted to KDD 2024) that proposed a formal position-weighted citation metric. See what GEO means.

Grounding query — The specific phrase or sub-query an AI system actually used when retrieving the content it cited — not necessarily the exact words the user typed. Bing Webmaster Tools' AI Performance report surfaces these directly so publishers can see what triggered a citation.

Hallucination — An AI-generated statement presented as fact that isn't supported by, or contradicts, the system's actual sources or training data. Relevant to brand accuracy: a hallucinated claim about a brand (wrong pricing, discontinued features, invented reviews) can appear with the same confident tone as an accurate one, and the reader has no visible way to tell the difference.

LLMO (LLM Optimization) — The broadest of these labels: shaping how any large language model describes or recommends a brand, in a search context or a plain chat conversation. Overlaps heavily with GEO but isn't limited to search-triggered answers.

Mention — The simplest and lowest bar: a brand's name appears anywhere in an AI answer's text, with or without a link or endorsement. See mentions vs citations vs recommendations.

Position — Where a brand appears when named in an AI answer, with 1 as first. One of ContextOwl's four fixed metrics. Position is also the least reliable of the four, since repeated runs of an identical prompt rarely return brands in the same order — see how many runs you actually need.

RAG (Retrieval-Augmented Generation) — The core technique behind most AI-search answers: the system retrieves relevant pages (using a ranking or search index) and then generates a response grounded in what it retrieved, rather than answering purely from what it memorized during training. Google's guide to its AI features names this explicitly as the mechanism behind AI Overviews and AI Mode.

Recommendation — An AI answer actively positioning a brand as a good fit for the question asked, not just present in a list. Can appear with or without a citation. See mentions vs citations vs recommendations.

Sentiment — Whether a brand is described positively, neutrally or negatively when named in an AI answer. One of ContextOwl's four fixed metrics; checked by reading the actual answer text, not inferred from a score.

Share of voice — A brand's slice of every brand mention across a fixed set of prompts and runs. One of ContextOwl's four fixed metrics; breaks if the prompt set or engine mix changes between measurements. See how to calculate it.

Visibility (rate) — How often an engine names a brand, measured across repeated runs of a fixed prompt set — not a single check. One of ContextOwl's four fixed metrics, and the one this glossary treats as the honest starting unit over position or a single bare score.

"AI visibility score" — Not a standardized metric. Any vendor's single bare score (including ContextOwl's own) should disclose what it's built from — which prompts, which engines, how many runs, which date — before it's treated as more than a quick read. See why one number misleads.

llms.txt — A proposed markdown file at a site's root meant to give AI agents a curated, LLM-friendly map of a site's content. Evidence that major AI search crawlers actually fetch and use it is thin; Google's guide states explicitly that Google Search doesn't use it. See whether it's worth bothering with.

SEO (Search Engine Optimization) — The established discipline of improving a page's visibility on a ranked results page. Google's own guidance treats optimizing for its generative AI features as an extension of SEO fundamentals rather than a separate discipline; other engines haven't published an equivalent statement.

Zero-click (search) — A search or AI-answer interaction where the person gets what they need without clicking through to a source page. Not a single agreed-on percentage — Pew Research Center measured an 8% click rate when a Google AI summary appeared versus 15% without one, a different measure than click-through-rate studies that track a specific page's ranking position. See whether SEO is actually dead because of this.

Which terms are synonyms and which are not?

Not synonyms, despite common interchangeable use: mention, citation and recommendation are three separately countable events, not three words for the same thing. Also not synonyms: visibility rate (how often you're named, across runs) and position (where you land when named) — a brand can have strong visibility and weak, inconsistent position at the same time.

Closer to true synonyms in casual use, though with real distinctions above: GEO, AEO and LLMO — three labels for overlapping work on slightly different surfaces. Most teams can pick one as an internal umbrella term without losing anything important, as covered in the full label comparison.

Which terms are contested?

llms.txt — contested. The proposal has real adoption among sites publishing the file, but Google states plainly it doesn't use it for Search, and independent crawler-traffic analysis has found AI bots rarely fetch it in practice.

Structured data / schema for AI visibility — contested. Google describes schema as part of general SEO practice but explicitly "not required," without claiming it improves AI-feature inclusion specifically.

"AI visibility score" as a single number — contested by design on this blog. We publish one, and disclose why it should be read as a snapshot, not a precise measurement.

FAQ

What's the difference between GEO and AEO? GEO specifically targets being quoted or cited inside a generative, LLM-based answer. AEO is the older, broader term covering any single-answer surface, including non-generative ones like classic featured snippets. Many teams use them interchangeably in practice.

Is "visibility rate" the same as "share of voice"? No. Visibility rate is how often you're named, on its own. Share of voice is your slice of all brand mentions relative to competitors across the same prompts — a comparative measure, not a standalone frequency.

Does ContextOwl use any metric not defined here? No. We use these four — visibility, share of voice, sentiment, position — consistently, and don't introduce new named scores beyond them.

Will more terms be added to this page? Yes. This is a living reference; new entries link out to a full post where one exists, and the page is reviewed every six months.

What to do next

1. Bookmark this page rather than a specific post — it's the fastest way to check which label or metric to use in your own reporting.
2. If you're choosing one internal term for client or leadership reporting, start with GEO vs AEO vs SEO vs LLMO rather than guessing from this list alone.
3. Each entry links to the post that covers it in depth where one exists — follow that link before quoting a definition in a client-facing document.

Check one of these terms against a real answer with the AI Citation Snapshot — ask one question, see whether your brand is named or cited, free.