Glossary of AI visibility terms
This glossary defines the terms used to measure how AI assistants talk about a brand, product, person or topic, and why. Each definition is one or two plain sentences, followed where useful by how Tonecast calculates it. Tonecast is a SaaS platform that shows what AI assistants say about a subject and which web pages shaped those answers.
- AI Overviews
- Google AI Overviews are AI-generated summaries that Google Search shows for some queries, with links to the pages they draw on. They are one of the AI answer surfaces Tonecast is adding to the engines it samples.
- AI visibility rate
- The AI visibility rate is the share of sampled AI answers in which the subject is mentioned. Tonecast reports it per engine, prompt, prompt category, language and country, with a 95% confidence interval.
- Campaign
- A campaign is the unit of work in Tonecast: one subject (a brand, product, public person or topic), the prompts asked about it, the languages and countries tracked, and a time window. A campaign does not define sources; the source graph is built automatically.
- Citation
- A citation is a link to a web page that an AI assistant includes in its answer as a source. Citations are the visible part of what an engine read; they do not list everything that shaped the answer.
- Cited sources
- Cited sources are the pages that the sampled AI answers cite explicitly. They form the Cited tier of the source graph, and their weight is certain: it comes from how often each engine cites them.
- Claim
- A claim is a factual statement about the subject, such as a price, a feature, a date or a location, found in an AI answer or in a source page. Tonecast extracts claims in a structured form: subject, property, value and the original sentence.
- Claim check
- Claim check is the comparison of the claims AI assistants make about a subject with its fact sheet. Each claim is classified as consistent, incorrect, outdated or unverifiable, and linked to the cited pages that carry it. Claim check reports and documents; it does not generate corrective content.
- Claim prevalence
- Claim prevalence is the share of the source graph, weighted by influence weight, that supports a given claim over time. A claim that grows in the sources before it appears in the answers is one of the early signals Tonecast watches.
- Confidence interval
- A confidence interval is the range within which a measured rate is likely to lie, given how many answers were sampled. Tonecast shows 95% intervals: Wilson score intervals for rates, and an interval on the mean for Tone in AI.
- Engine
- An engine is an AI assistant or AI answer surface that Tonecast samples: ChatGPT, Claude, Gemini and Perplexity today, queried through their official APIs, with Google AI Overviews on the way.
- Fact sheet
- A fact sheet is the list of verified facts about a subject that its owner maintains in Tonecast: prices, features, dates, locations, official statements and what the subject is not, each with a validity date. It is the reference for claim check.
- Formative Sentiment Index (FSI)
- The Formative Sentiment Index is the net tone of the pages in a subject's source graph, weighted by their influence weight, on the same −100 to +100 scale as Tone in AI. It shows whether the sources that shape AI answers are becoming more positive or more negative.
- Formative sources
- Formative sources are pages that AI engines are likely to draw on for a subject even when they do not cite them: discussion threads, encyclopedia entries, videos, news articles, reviews and comparison lists. They form the Formative tier of the source graph, and their influence is estimated.
- Generative Engine Optimization (GEO)
- Generative Engine Optimization is the practice of making a brand, product or piece of content accurately represented, cited and recommended in the answers of AI assistants and AI search features. It is also called AI search optimization or answer engine optimization (AEO).
- Influence weight
- An influence weight is Tonecast's estimate of how much a page is likely to shape a given engine's answers about a subject. It combines citation frequency, how much the engine typically draws on that kind of source, search position for the prompts and recency; its components are stored and it is re-estimated every week.
- Lead indicator (early signal)
- The lead indicator, shown as Early signals, tests whether changes in the sources come before changes in the AI answers, for each engine. It has three states: none, observing and active. It becomes active only when the lead is statistically significant after correction for multiple testing; only then does it produce an early warning.
- Lead time
- Lead time is the estimated number of days by which a change in the sources precedes a change in an engine's answers. Tonecast reports it only for active early signals.
- Methodology card
- A methodology card is the page, one per campaign, that states how its numbers were produced: prompts, languages, engines, number of samples, sampling window, confidence intervals, measured model accuracy against quality targets, and the state of the early signal.
- Owned property
- An owned property is a website, page or official profile that belongs to the subject and that its owner declares in Tonecast, such as the company site, product pages or FAQ. Declaring it separates your own pages from the pages others publish, and lets claim check trace outdated claims back to your own pages.
- Owned sources
- Owned sources are the pages of the declared owned properties. They form the Owned tier of the source graph and are read automatically.
- Recommendation rate
- The recommendation rate is the share of AI answers mentioning the subject in which the engine explicitly recommends it. Being mentioned is not the same as being recommended.
- Sample
- A sample is one run of one tracked prompt on one engine, stored in full with its citations. Rates are computed over many samples, because answers vary between runs.
- Sampling cadence
- The sampling cadence is how often the tracked prompts are sent to the engines. Tonecast samples every week, with a number of runs per prompt, per engine, per month set by the plan (6 on Essential, 8 on Starter, Pro and Agency).
- Share of voice in AI answers is the subject's mentions as a share of all mentions of the subject and its declared competitors in the sampled answers. It shows who the engines talk about when they answer questions in your category.
- Source graph
- A source graph is the set of web pages on which what AI assistants say about a subject was formed, organised in three tiers: Cited, Formative and Owned. In Tonecast it is built automatically for each campaign, and every page records the reason it was included and its influence weight. The view is called What shaped it.
- Tone in AI
- Tone in AI is the net tone of what AI answers say about the subject, from −100 (all negative) to +100 (all positive), measured towards the subject rather than on the whole text. It is computed over the answers that mention the subject.
- Tracked prompt
- A tracked prompt is a question that Tonecast asks the AI engines on a schedule, in one language, such as "best compostable coffee capsules" or "is this brand reliable?". The same question in two languages counts as two tracked prompts.