GEO & AI visibility glossary

LLM citations are sources, links, or attributions an AI answer shows when it grounds a claim in retrieval or browsing. In visibility audits, they also mean how often a model names your brand with enough context to count as a substantiated mention on buyer prompts.

→ View Agencies benchmark

Brand recall rate in AI contexts is how often a large language model names your brand unprompted or in the first position when users ask category or problem questions — compared to how often it names competitors for the same intents. It differs from share of voice, which counts any mention; recall emphasizes top-of-mind recommendation strength. For ecommerce, high recall means your catalog and positioning show up when shoppers ask for product types or use cases, not only when they search your name.

→ View E-commerce benchmark

An AI visibility audit is a structured test of how large language models describe and recommend your brand when given realistic buyer questions — usually across several models and query variants. It produces evidence of mention frequency, position, tone, and consistency rather than a single anecdotal chat. Startups use it to see whether early positioning and launch coverage actually register in AI answers before spending on broad SEO or paid channels.

→ View Startups benchmark

Recommendation engine optimization is the practice of improving how often and how favorably your brand is surfaced in algorithmic recommendation contexts — including AI assistants that suggest tools, vendors, or products in natural-language answers. It overlaps with GEO but emphasizes ranking-like ordering inside a generated answer list rather than crawl or keyword position. For SaaS, it matters when buyers ask “what should we use for…” and the model returns an ordered or ranked set of names.

→ View SaaS benchmark

AI search share of voice is the share of AI-generated or AI-mediated search-style answers in your category that reference your brand versus competitors — including answers that blend retrieval with model reasoning. It is broader than classic web search impressions because many answers never produce a click. For B2B, it indicates whether committees see your name when they ask discovery questions in tools that summarize or recommend vendors.

→ View B2B benchmark

An LLM training signal is any text pattern in pre-training or fine-tuning data that teaches a model to associate entities — brands, products, problems — with each other. Strong signals come from repeated, consistent mentions across independent sources; weak or contradictory signals produce uncertain recommendations. Agencies explaining AI visibility to clients use this term to separate “we said it on our site” from “the model learned it from many places.”

→ View Agencies benchmark

A zero-click AI answer is a complete response from a large language model or AI search surface that satisfies the user without a follow-up link click — the user gets names, steps, or comparisons entirely in the chat or answer panel. For brands, it means your discovery and differentiation must be expressible in what the model says aloud, not only on your landing page. Ecommerce brands feel this when shoppers add to cart from a recommendation list without visiting SEO snippets.

→ View E-commerce benchmark

A generative answer engine is a system that produces natural-language answers to user questions — combining retrieval, web context, and large language model generation — as opposed to returning only ten blue links. Examples include AI modes in search products and standalone assistants. Startups need to treat these engines as primary discovery surfaces because users may never see a traditional results page.

→ View Startups benchmark