Share of Model FAQ & Glossary: Every Term Explained
AI search introduced a new vocabulary almost overnight, and much of it gets used loosely — “share of voice” and “share of model” are still confused in vendor decks, and terms like “citation” and “mention” get swapped even though they mean different things inside an answer engine. This glossary defines the terms that matter for measuring and improving AI visibility, grouped by theme, so you have one accurate reference instead of piecing definitions together from ten different blog posts.
Each entry is written the way we’d explain it to a client on a call: plain language first, precision second. Bookmark this page as the reference companion to the rest of this cluster on measuring share of model.
Core Visibility Metrics
Share of Model
The percentage of relevant AI-generated answers, across a defined set of prompts, in which your brand, product, or content appears — whether as a citation, a mention, or a direct recommendation. It is the AI-search equivalent of share of voice, but measured against language model outputs instead of search engine results pages.
Share of Voice
A pre-existing marketing metric describing how much of the total conversation, media coverage, or search visibility in a category belongs to your brand versus competitors. In traditional SEO, it’s typically calculated from keyword rankings and estimated traffic. Share of model borrows this logic but applies it to answers generated by AI systems rather than ranked links.
AI Visibility
The umbrella term for how often, how prominently, and how favorably a brand appears across AI-driven answer surfaces — AI Overviews, chatbots, AI-powered search assistants, and other generative interfaces. Share of model is one way to quantify AI visibility; sentiment and recommendation share are others.
Recommendation Share
A narrower, higher-value subset of share of model: the percentage of prompts in which a model doesn’t just mention your brand in passing, but actively recommends it as a solution, option, or best pick. A brand can have decent share of model but weak recommendation share if it’s cited as background context rather than suggested to the user.
Benchmark / Baseline
The first measurement taken before any optimization work begins, used as the point of comparison for all future tracking. Without a documented baseline — the exact prompt set, the date, the models tested, and the raw results — it’s impossible to prove that share of model has actually improved rather than simply appearing to, due to model updates or prompt drift.
How Models Reference Content
Citation
An explicit reference a model makes to a specific source — typically shown as a linked footnote, a named source, or an inline attribution in AI Overviews, Perplexity, or similar retrieval-augmented systems. Citations are trackable and clickable, which makes them the closest AI-search equivalent to a ranked organic listing.
Mention
An appearance of your brand, product, or entity name within an AI-generated answer that is not necessarily tied to a clickable source link. A model can mention a company by name because it learned about it during training, without citing any specific page. Mentions matter for brand presence even when they don’t drive a click.
Attribution
The broader act of a model crediting information to a source, whether through a formal citation, an in-text mention, or a paraphrased reference (“according to industry data…”). Attribution quality — how accurately and fairly a model represents your content — is as important to track as attribution frequency.
Sentiment
The tone or valuation attached to a mention: positive, negative, or neutral. A brand can achieve strong share of model numerically while carrying negative sentiment (e.g., being cited as a cautionary example or “budget” option), which is why sentiment should always be tracked alongside raw appearance frequency, not as an afterthought.
The Systems Doing the Answering
LLM (Large Language Model)
A machine learning model trained on massive amounts of text to predict and generate language. LLMs power chatbots like ChatGPT, Claude, and Gemini, and increasingly sit behind traditional search results in the form of AI-generated summaries. Understanding that an LLM generates probabilistic text — not retrieved facts by default — is foundational to understanding why hallucination happens.
RAG (Retrieval-Augmented Generation)
An architecture where a model first retrieves relevant documents from a live index or database, then generates its answer grounded in that retrieved content. AI Overviews, Perplexity, and most AI search assistants use some form of RAG, which is why traditional SEO fundamentals — crawlability, clear structure, authoritative content — still influence whether you get retrieved (and therefore citable) in the first place.
AI Overviews
Google’s AI-generated summary that appears above traditional organic results for many queries, synthesizing information from multiple sources into a single answer with citations. Appearing in an AI Overview is a distinct visibility event from ranking organically, and it’s one of the most commonly tracked citation surfaces in share of model measurement.
Answer Engine
A general term for any AI system whose primary job is to directly answer a user’s question rather than return a list of links to click through. Search engines are becoming answer engines; the shift is central to why share of model exists as a metric at all.
Prompt Set
The defined, documented list of prompts (questions or queries) used consistently to test AI visibility over time. A well-built prompt set mirrors real customer language — informational, comparison, and transactional phrasings — rather than just your target keywords rewritten as questions. Consistency in the prompt set is what makes benchmark comparisons valid.
Search Behavior Concepts
Zero-Click
A search or prompt interaction in which the user gets their answer directly from the results page or AI-generated response without clicking through to any website. AI Overviews and chatbot answers have accelerated zero-click behavior, which is why measuring visibility inside the answer itself — not just downstream clicks — has become necessary.
Query Fan-Out
The process by which an AI system breaks a single user prompt into multiple related sub-queries or retrieval passes to gather broader context before generating its final answer. This means a page can influence an AI answer even if it doesn’t perfectly match the user’s literal wording, because the system fanned out to related phrasings your content happens to cover.
Conversational Search
Search behavior conducted through multi-turn, natural-language dialogue rather than isolated keyword queries — the interaction style native to chatbots and AI assistants. Content written to answer a single, narrow question well tends to perform better in conversational search than content written to rank for a keyword phrase.
Content & Entity Fundamentals
Entity
A distinct, identifiable thing — a person, brand, place, product, or concept — that a model or search engine can recognize and reason about independent of specific wording. Being understood as a well-defined entity (rather than just a string of text that happens to match a query) is a prerequisite for consistent, accurate AI mentions.
Knowledge Graph
A structured database of entities and the relationships between them, used by search engines and increasingly by AI systems to verify and connect facts. Strong, consistent entity signals across the web (structured data, consistent naming, authoritative third-party references) feed into how confidently a model represents your brand.
Structured Data / Schema Markup
Machine-readable code added to a webpage (commonly in JSON-LD format) that explicitly labels content elements — author, organization, FAQ, product details — for search engines and AI crawlers. Structured data doesn’t guarantee a citation, but it removes ambiguity that could otherwise cause a model to misattribute or skip your content during retrieval.
Hallucination
An instance where a model generates information that sounds plausible but is factually incorrect, fabricated, or not actually supported by any real source. Hallucinations are a core risk in share of model tracking, because a model can “mention” your brand inaccurately — misstating your services, pricing, or founding date — which is a distinct problem from simply not being mentioned at all.
E-E-A-T
Google’s framework — Experience, Expertise, Authoritativeness, Trustworthiness — for evaluating content quality, originally built for human search quality raters but increasingly relevant to what AI systems favor when selecting sources to cite. Content demonstrating real, first-hand experience tends to be retrieved and cited more reliably than generic, templated material.
Frequently Asked Questions
Is share of model the same thing as share of voice?
No. Share of voice traditionally measures presence across search rankings, media coverage, or social conversation. Share of model measures presence specifically within AI-generated answers, across a defined prompt set. The concepts are related, but share of model requires different tracking methods since there's no ranked list to scrape — you have to run the prompts and analyze the actual generated text.
What's the difference between a citation and a mention?
A citation is an explicit, usually clickable, attribution to a specific source within an AI answer. A mention is your brand or entity name appearing in the answer text without necessarily linking back to a source — the model may simply "know" about you from training data. Both count toward AI visibility, but citations are more directly actionable and trackable.
Why does a model sometimes get facts about my business wrong?
This is typically hallucination, or it's the model relying on outdated or low-authority sources during retrieval. It's more likely when your official information is inconsistent across the web, when your site lacks structured data, or when third-party sources about your business are sparse or contradictory. Consistent, well-structured, frequently updated content reduces the odds of misattribution.
How often should I re-run my prompt set to track share of model?
There's no universal cadence, but monthly or quarterly re-testing is common practice, since it balances catching meaningful shifts against the cost and noise of testing too frequently. What matters more than frequency is consistency: use the same prompt set and the same evaluation criteria every time, so changes reflect real shifts in visibility rather than measurement variance.
Does ranking well in traditional Google search guarantee AI visibility?
No. Ranking and being retrieved are related but separate mechanics. Many RAG-based systems pull from their own indexes or a mix of sources, and a page can rank highly in organic search while never being surfaced in an AI answer, or vice versa. Strong fundamentals — clear structure, authoritative signals, accurate structured data — help with both, but they aren't interchangeable outcomes.
What is a "recommendation" versus just a "mention" in an AI answer?
A mention is passive: your brand name appears somewhere in the response, often as background context. A recommendation is active: the model presents your brand as a suggested option or best choice for the user's need. Recommendation share is generally the more valuable metric to optimize toward, since it reflects the model treating your brand as a credible answer rather than incidental context.
Terry has 30+ years in software and SEO. He’s the founder of Salterra Digital Services and SEO Spring Training, host of the Roundtable SEO Mastermind, and lead instructor at SEO University — teaching the exact tactics his team uses on client work.
Continue learning
Ready to master this?
This guide is one lesson from the Measuring AI Visibility Share of Model course. Get every lesson, framework and checklist — plus the full 38-course catalog — inside SEO University.
Practitioner-focused training across the full digital marketing stack — from technical SEO to conversion optimization and the AI search era. By Salterra Digital Services, since 2011.