Resources / Glossary

The GEO vocabulary, defined plainly.

AI search has its own language, and most of it is less than two years old. Here are the 29 terms we use across this site and the product, grouped from the core idea out to the words specific to how AuthorityOn scores a brand.

The core idea

Why a new vocabulary exists at all.

Generative engine optimizationGEO: The practice of improving how a brand is represented in the answers generated by AI assistants, the way SEO improves how a site ranks in a list of links.
AI search: The pattern where a buyer asks an assistant like ChatGPT, Claude, Gemini or Perplexity a question and reads a synthesised answer, instead of clicking through a ranked list of links.
Answer engine: A system that responds to a query with a single composed answer rather than a page of sources. It is the surface a brand now competes inside.

How models answer

The mechanics that decide whether a brand shows up.

Retrieval-augmented generationRAG: A method where the model retrieves live documents at the moment of the question and writes its answer from them, rather than from training memory alone.
Grounding: When a model's answer is tied to retrieved, citable sources rather than produced from its internal training weights. Grounded answers are where source attribution matters most.
Knowledge cutoff: The date after which a model has no training data. Anything published later is invisible to it unless retrieved live at query time.
Retrain event: A model version update. It can change a model's answers across the board overnight, which is why what AI says about a brand has to be watched, not measured once.
Hallucination: A confident, fluent model claim that no real source supports. For a brand, the risk is a wrong fact stated plainly to a buyer.

What we measure

The signals a scan puts numbers on.

Spontaneous recommendation: When a model names a brand in answer to a category question without the buyer having mentioned that brand. The rate at which this happens is the clearest single read on AI visibility.
Head-to-head win rate: How often a model picks a brand when asked to compare it directly against a named competitor. A brand can win every head-to-head and still almost never be recommended unprompted.
AI share of voice: The proportion of category answers a brand appears in at all. The denominator is every relevant buyer question, not just the ones that name the brand.
Source attribution: The set of third-party pages and publications a model draws on to answer a category question. Earning placement in those sources is how a brand changes the answer.
Drift: Change over time in what models say about a brand, as they retrain, re-rank sources, or the underlying web changes.
Polarity: Whether a brand mention is positive, neutral, or critical. The same level of visibility reads very differently depending on its polarity.
Prompt set: The bank of buyer-style questions a scan runs against each model. AuthorityOn runs 200+ per scan, spanning the questions a real buyer would ask.
Brand mention detection: Identifying whether, where, and how a brand appears inside a model's response, so visibility can be counted rather than guessed.

Crawling and indexability

Whether the models can reach a site in the first place.

AI crawler: An automated agent that fetches web pages for a model's training corpus or its live retrieval index. If a crawler cannot reach a site, the model cannot learn from it.
GPTBot: OpenAI's web crawler. Site owners allow or block it in robots.txt.
ClaudeBot: Anthropic's web crawler, the agent that fetches pages for Claude.
PerplexityBot: Perplexity's crawler, which gathers pages for its answer engine.
Common Crawl: A large, openly published crawl of the web that many models draw on as training data. Blocking it removes a site from a corpus used industry-wide.
Google-Extended: The control that tells Google whether a site may be used for Gemini and other Google AI products, separate from ordinary Google Search indexing.
llms.txt: A proposed file, placed at the root of a site, that points AI crawlers at the content most worth reading. A complement to robots.txt, not a replacement.
Indexability: Whether a model's crawler can actually reach and read a site. A site can have excellent content and still be invisible because one firewall rule blocks the crawlers.

AuthorityOn vocabulary

Terms specific to how AuthorityOn scores and tracks a brand.

Signal Five: AuthorityOn's five-part read on AI authority: spontaneous recommendation, head-to-head win rate, discovery and indexability, source attribution, and drift.
Composite signal score: A blended score across the measured signals, scored separately for every model so a brand can see where it stands per assistant.
Brand and entity: A brand is a top-level thing AuthorityOn tracks. An entity is a thing tracked under it: a product, an executive, a campaign, a sub-brand. The two are counted as separate limits.
Citable page: A page written so a model can lift a clean, attributable claim from it. Citable pages are how a brand earns its way into the answer for a category question.
Basic and advanced scan: A basic scan returns the full score and findings. An advanced scan adds the generated recommendation layer: the prioritised action plan on top of the diagnosis.

See these signals scored on your brand.

Your first audit lands the moment you sign in: the Signal Five scored, every model separately, and a prioritised action plan. From €49/mo, cancel any month.

Start subscription Back to resources →