Glossary of every term

Every term the tool uses, defined once. If a screen ever uses a word you do not recognize, it is here. Ordered roughly from most-common to most-technical.

The core numbers

Citation share The percentage of tracked question-and-engine combinations where an AI answer cited your website as a source on a majority of replicates. Your headline visibility number. Objection questions are excluded from it.

Mention rate How often an answer named your brand at all, whether or not it linked you. You can have a high mention rate and low citation share, which is a specific, fixable situation (people talk about you; engines source their facts elsewhere).

Share of voice (SOV) Of all the brand mentions across your tracked answers, what share are yours versus your tracked competitors. Measures how much of the conversation you own.

Average rank When several brands are named in one answer, your ordinal position among the tracked brands (first-mentioned, second, and so on). Lower is better. It is a position among tracked entities, not a word-count offset.

How measurement works

Engine One of the four AI answer products the tool queries: ChatGPT, Claude, Gemini, Perplexity. The tool also tracks Google’s two AI search surfaces, Google AI Overviews and Google AI Mode, alongside them. Numbers come from each engine’s API, which is a proxy for the consumer product, not identical to it.

Replicate One repeat of a question on one engine. Engines are non-deterministic, so the platform can run each question multiple times; the pass count per run is a platform setting, and each run’s results show the count that actually ran. Whenever more than one pass runs, a result only counts if it appears in the majority of replicates.

Panel Your full set of tracked questions, organized by category, persona, funnel stage, and sub-topic. You manage it in Questions.

Check / run One execution of the panel across the engines. Runs are scheduled (the weekly cron) or on demand. Each run is what produces a fresh Scorecard.

Baseline A deeper first run used to establish a starting point.

Evidence and honesty

OBSERVED A finding the tool directly measured, with a traceable source reference. Fact.

INFERRED A professional judgment based on the evidence, clearly labeled as such. Expert opinion, not measurement.

Gate / gate-blocked A foundational problem that makes everything downstream moot until it is fixed. The gates are T1/T2/T3 on a page (AI crawlers cannot reach the page, its content is not in the raw HTML, or its structured data is not in the raw HTML, so non-rendering crawlers cannot see it) and S1/S5 on the site (the homepage serves no server-rendered HTML to a plain fetch, or the site tells crawlers not to index or quote it with noindex/nosnippet). A failed gate caps its category’s score, so a blocked layer never reads as passing. A gate-blocked score is the tool refusing to grade content nobody can read, not hiding your score.

Instrumented / not instrumented Whether the data source behind a number is connected yet. “Not instrumented, no Cloudflare zone” means crawler analytics is blank because Cloudflare is not connected, not that the value is zero.

Insufficient window Trend and drift alerts need eight weeks of history to fire reliably. Until then the tool stays quiet (“2 of 8 weeks”) rather than alerting on noise.

Answer-level terms

Ghost citation An answer that cites your site as a source but never names your brand in the text. The reader gets your information without learning it came from you. Surfaced on the Scorecard and filterable in Answers.

Query fan-out The web searches an engine ran on its own while answering, parsed from its raw response. Shows the exact phrasings the engines think your buyers use. Some engines expose these (ChatGPT, Claude, Gemini); Perplexity does not, and the tool says so.

Word transformation A word an engine added when rewriting your question into its own search, that was not in your original prompt. A signal of the angles the models believe your buyers care about.

Decision journey (TOFU / MOFU / BOFU) Top, middle, and bottom of funnel. Your questions are grouped by how close the buyer is to a decision, so you can see whether you show up during research versus at the moment of choice.

Content and fixes

Extractability How easily an AI engine can lift a clean, correct, self-contained fact from your page. High-extractability content gets cited more. Studio has a scorer and a rewriter for it.

Fact sheet The approved set of true statements about your brand, each with a source. Once you approve it in Brand profile, the tool checks AI answers against it (factuality checking). Until approved, factuality checking stays off.

llms.txt A curated file (like robots.txt, but for AI) that tells engines what your brand is and points them at your best pages. Studio drafts one in the recommended “Pattern C” style.

Anchor shift A content strategy for when engines hold a negative frame about you: rather than rebutting it (which reinforces it), you introduce a stronger adjacent frame where you win. Studio generates these.

Refresh tier How often a published page should be refreshed to stay citable (for example 90 days, 6 months, or 60 days for sensitive YMYL topics). Managed in Studio’s editorial calendar.

Layers

The tool organizes measurement into four layers you will see referenced:

1. Layer 1 — what the AI answers say (the panel).
2. Layer 2 — which AI crawlers actually visit your site (needs Cloudflare).
3. Layer 3 — your own page’s AEO readiness (the audit).
4. Layer 4 — business outcome (AI-influenced revenue; a later phase).