- AI visibility
- How often, in which position and with what tone a brand appears in the answers generated by AI assistants. Unlike a search ranking it has no page two: a brand is named inside the answer or it is absent from the decision.
- Answer engine optimisation (AEO)
- The practice of getting a brand named inside AI-generated answers rather than ranked on a results page. The work is mostly done on third-party sources the model trusts, not on pages you own.
- Generative engine optimisation (GEO)
- Used interchangeably with AEO. Where the two are distinguished, GEO leans toward shaping the generated text itself — the claims and phrasing a model reproduces — rather than only whether your name appears.
- Brand mention
- Your brand's name appearing in the text of an answer. It is what determines whether you are recommended, and it is measured separately from whether your content was used.
- Source citation
- The answer being built from your content, with or without your name in the text. A page can feed dozens of answers while the brands named in them are competitors.
- Share of voice
- Your proportion of all brand mentions across a tracked set of category prompts, over a period. It is the closest AI-search equivalent of category market share.
- Prompt
- One tracked question, in one language and one market. Prompts are the unit of measurement and, on most platforms including Rixe, the unit of pricing.
- Query fanout
- The model silently breaking one question into several sub-queries, retrieving for each, then composing a single answer. It is why a brand can be strong on the visible question and absent from the answer.
- Grounding
- The retrieval step that lets a model read the live web before answering. Without it a model answers from training data alone and will name brands that no longer exist.
- AI Overviews
- Google's generated summary above the classic results page. It behaves differently from a chat assistant: it leans harder on sources it can link, which makes citations matter more than prose mentions.
- Tone score
- A rating of the sentences built around your name, separate from whether you appear at all. The same visibility score reads very differently at a tone of 37 than at 81.
- Confidence interval
- The band around a measured score that says how certain it is. Generative answers vary between runs, so a score built from twelve observations deserves a wider band than one built from ninety.
- Answer surface
- The place a generated answer is shown — a chat reply, an AI Overview above the results, an assistant inside a browser. Each one retrieves differently, so a brand can be present on one and absent from another.
- Retrieval-augmented generation (RAG)
- Answering by fetching documents first and writing from them, rather than from the model’s memory alone. It is why a page you publish today can appear in an answer tomorrow without the model being retrained.
- Crawler (AI)
- The bot that fetches your pages so an engine can quote them. It is not the same bot as a search crawler, it follows its own robots rules, and blocking it removes you from the answers regardless of how you rank.
- Answer format score
- How liftable a page is: a direct answer near the top, a question as a heading, short self-contained paragraphs, a table where a table belongs. Accuracy and readability are separate qualities from this one.
- Prompt set
- The defined list of questions a brand is measured on. A visibility number is only as meaningful as the set behind it, which is why the set is stated rather than assumed.
- Pass rate
- The share of runs on a prompt where the brand was named. Because engines are non-deterministic, a prompt is run repeatedly and the pass rate — not a single answer — is the measurement.
- Non-determinism
- The same prompt producing different answers on different runs. It is a property of the engines, not a fault in the measurement, and it is why a score is published with a confidence interval and a run count.
- Coverage
- Whether a prompt set spans the whole buying journey — awareness, consideration, evaluation, purchase — or crowds into one stage. A gap here means questions nobody is measuring, and usually nobody is answering.
- Ad load
- How much of an answer surface is paid placement: whether the page carried an ad, how many, and which of them sat inside the generated answer rather than around it.
- AI referral
- A visit that arrived from an AI answer rather than a results page. It is read from your own Search Console and analytics, not modelled, and it is what ties a mention to something that actually happened.
- Log drain
- A stream of your server or CDN access log sent to a measurement tool. It is how you see which AI crawler fetched which URL and when, instead of estimating how often you were probably crawled.
- IndexNow
- A protocol for telling search indexes that a page changed, instead of waiting to be recrawled. It matters for Bing in particular, because ChatGPT and Copilot answer from that index.
- Entity
- The thing a model believes your brand is — a name attached to a category, a set of attributes and a set of relationships. Most wrong answers are an entity problem before they are a content problem.
- Hallucination
- A confident statement with nothing behind it. In brand terms it is usually not invention from nothing but an old fact repeated: a price, a limit or a feature that was true two years ago and is still being quoted.
+2
