AI visibility glossary: 18 terms in plain English

17 August 2026 · AI Visibility Index

Why this exists: the AI-search industry sprouts jargon faster than it sprouts standards. Half of it describes real mechanics; half is marketing fog. Here's the vocabulary you'll meet, in plain English, with an honest note on what actually matters.

The big ideas

AI visibility — How often AI engines name your business when people ask buying questions. Measured as a rate across many prompt runs, not a single ranking. The metric that matters most.

GEO (Generative Engine Optimisation) — The practice of improving how generative AI engines represent and recommend you. The AI-era sibling of SEO.

AEO (Answer Engine Optimisation) — Older term for the same idea, from the featured-snippet era. Now mostly folded into GEO.

Share of model — Of all the times an AI recommends someone in your category, the share that names you. Like share of voice, for AI answers.

Entity — A thing the model recognises as real and distinct: a business with a name, place, and consistent identity. Becoming a clean entity is half the battle — models recommend things, not strings of text.

Citation — When an AI answer links to or names a source. Being cited (as a source) and being recommended (as an option) are different wins; you want both.

The machinery

AI crawler — The bot that fetches your pages for an AI company. The ones to know: GPTBot (OpenAI training), OAI-SearchBot (ChatGPT search citations), ChatGPT-User (live browsing), PerplexityBot, ClaudeBot, Google-Extended. Each can be allowed or blocked separately in robots.txt — our allow-list guide has the config.

Retrieval — The step where the AI fetches current information before answering. If retrieval can't find you, no amount of training-data presence saves you.

RAG (Retrieval-Augmented Generation) — The architecture behind search-enabled AI: retrieve documents, then generate an answer grounded in them. Why "being findable" and "being quotable" are two separate problems.

Index — The database of crawled pages an engine searches. Google AI uses Google's index; ChatGPT leans on Bing's; Perplexity runs its own.

Grounding — Anchoring an answer in retrieved sources rather than the model's memory. Grounded answers are the ones that can name you correctly.

Hallucination — When a model states something false with confidence. Clean entities and consistent data across the web are the best defence against hallucinated claims about your business.

The formats

llms.txt — A proposed briefing file at your site root telling AI models what you are and what matters. Plain-English guide here.

JSON-LD — The script format for schema markup. Machine-readable identity for your business — with examples.

Schema / structured data — Standardised tags (schema.org) describing businesses, articles, FAQs. The vocabulary JSON-LD carries.

Answer-first content — Pages that lead with a 75–150 word direct answer before the detail. The format AI engines extract most reliably.

Prompt tracking — Running fixed sets of buyer questions against AI engines over time to measure visibility. What our data engine does nightly.

Frequently asked questions

Is GEO just SEO with a new name?

They overlap heavily — good structure, clear entities, quality content help both. The differences are in retrieval sources (Bing matters again), output format (sentences, not links), and measurement (visibility rates, not positions).

Which single term should I care about most?

AI visibility rate — the share of relevant buyer questions where you get named. Everything else is a means to move that number.

Are "AI SEO tools" that promise rankings legit?

Be sceptical of anyone promising fixed positions in AI answers — answers are regenerated every run and vary. Legitimate tools measure patterns across many runs and show their evidence. That's the standard to hold vendors to, us included.

---

Originally published at the AI Visibility Index.