When a marketing team talks about "AI presence," everyone in the room pictures something different. One person thinks about ads inside ChatGPT. Another thinks about prompt engineering. A third assumes SEO already covers it. The disconnect starts with language.
This glossary is here to fix that. It's not a list of buzzwords to sound current — it's a working vocabulary so your team makes decisions based on the same definitions.
Terms are organized by area. Use this as a shared reference, not a reading assignment.
How LLMs actually work
LLM (Large Language Model)
A language model trained on large volumes of text to generate coherent responses. ChatGPT, Claude, Gemini, and Llama are all LLMs. Critical to understand: LLMs don't "search" — they predict the most likely text based on what they've learned.
Training data
The body of text used to train the model. Content that appears frequently and consistently in training data tends to carry more weight in responses. Your old blog posts, forum mentions, press coverage — all of it may have been included here.
Knowledge cutoff
The date beyond which the model has no training data. If your product launched after the cutoff, the AI simply doesn't know it exists — unless a real-time retrieval mechanism is in play.
RAG (Retrieval-Augmented Generation)
A technique where the model pulls external information at query time before generating a response. ChatGPT with browsing uses RAG. It checks live sources, brings in excerpts, and incorporates them into the answer. This creates a second window of opportunity for your brand to appear — beyond training data.
Grounding
The process of anchoring AI responses in verifiable sources. Gemini uses Google's index for grounding. When a model is "grounded," it's more likely to cite a real source — which is why search presence still matters even in an LLM world.
Embedding
A mathematical representation of text that captures meaning. Models use embeddings to compare concepts and decide what's "relevant" to a query. If your brand has weak semantic associations with a category, it may simply be skipped.
Hallucination
When the AI generates a factually incorrect statement with confidence. In a brand context, hallucination can mean wrong pricing, features that don't exist, or invented history. It's a real operational risk, not just an interesting quirk.
Visibility and presence
AEO (Answer Engine Optimization)
The practice of optimizing content to appear in the responses of answer engines — specifically the LLMs people consult directly. Unlike SEO, the goal isn't ranking in a list: it's being cited or referenced in a natural-language response.
GEO (Generative Engine Optimization)
A broader synonym for AEO, covering all optimization aimed at generative engines — LLMs with retrieval, Google's AI Mode, Perplexity, and so on. Some authors draw distinctions; in practice the terms are used interchangeably.
AI Visibility
The ability of a brand to be recognized, cited, or referenced by AI systems in relevant conversations. It's the umbrella concept: it covers frequency, context, sentiment, and accuracy of what AI says about you.
Share of voice in AI
The proportion of times your brand appears in AI responses compared to total brand citations in your category. It's a relative metric. If the AI mentions you in 3 out of 10 responses about your market, and a competitor appears in 7, you hold 30% share of voice.
Citation worthiness
The degree to which a brand, source, or piece of content is recognized as worth citing by an LLM. It depends on perceived authority, information consistency, and presence across sources the model considers trustworthy.
Brand mention
When an LLM includes your company name in a response — positive, neutral, or negative. Mentions need to be tracked for both frequency and context.
Metrics and monitoring
LLM Brand Monitoring
The systematic practice of tracking how and when your brand appears in language model responses. It includes logging mentions, assessing sentiment, checking factual accuracy, and benchmarking against competitors.
Monitoring prompt
A question — or set of questions — sent regularly to LLMs to audit how they respond about your brand, category, or competitors. The AI-era equivalent of the "search query" in SEO, but in the form of natural-language questions.
Localized prompts
Prompts adapted for language, cultural context, and local market. "Best project management software" may produce entirely different responses in US English versus Brazilian Portuguese — even within the same model.
Citation sentiment
The tone with which the AI mentions your brand: positive, neutral, or negative. A brand can be cited frequently while still being mentioned with caveats or errors that damage user perception.
Citation accuracy
How factually correct the AI's information about your company is. Pricing, location, product features, company history — every repeated error at scale is a real business problem.
Presence baseline
A snapshot of your current AI visibility before any interventions. Essential for measuring the impact of changes over time. Without a baseline, you're flying blind.
Content and strategy
Citable content
Content written with enough clarity, specificity, and assertiveness that an LLM will extract it directly as a source. Definitions, lists, comparisons, and proprietary data tend to be more citable than vague narrative prose.
Proprietary data
Original research, benchmarks, surveys, and datasets produced by the company itself. Highly citable because they offer something that doesn't exist anywhere else. An LLM looking for a specific number will favor a primary source.
Earned media
Coverage obtained on merit — articles, third-party blog mentions, independent reviews. In an AEO context, earned media distributes brand information across sources that LLMs tend to find more trustworthy than your own website.
Schema markup (structured data)
HTML code that organizes page information in a machine-readable format. Still relevant for LLMs that use retrieval, especially those drawing from a search index.
High-authority source
A publication, encyclopedia, or platform that LLMs treat as inherently trustworthy. Wikipedia, established industry publications, and large-audience portals tend to have disproportionate weight in citations.
User behavior and the funnel
Zero-click
When a user gets the answer directly from the AI without clicking any link. Traffic doesn't reach your site, but your brand may still have been mentioned. Zero-click isn't inherently a loss — it depends on where that interaction sits in the journey.
AI Mode (or Search Generative Experience)
A search mode that uses generative AI to synthesize a response above traditional results. Google's AI Mode is the evolution of the SGE. It fundamentally changes who appears and how.
AI agent
An AI system that executes tasks autonomously — researching, comparing, deciding. Agents that make purchases or select vendors on a user's behalf create a decision layer where brands need to be present without the end user ever seeing the search happen.
AI funnel
The user journey that starts inside an LLM — with a question — and can reach a purchase decision without passing through the traditional flow (search → website → conversion). It requires rethinking where and how your brand shows up.
Shared vocabulary is the starting point. The next step is measuring where you actually stand — before committing to any tactic. Genoma was built for exactly that: auditing your current presence across the leading LLMs, comparing you to competitors, and surfacing where to act first. You can see how it works at app.genomahq.com.