What Tonecast measures, and how it explains it

AI Answers: what do AI assistants say about you?

AI Answers is the Tonecast view of what AI assistants actually answer when people ask about your subject. Every week Tonecast sends your tracked prompts to each engine through its official API, with web search or grounding enabled, and keeps every answer in full so that each number can be opened down to the answers behind it.

For every engine, prompt, prompt category, language and country, AI Answers reports:

• AI visibility rate: the share of answers in which your subject appears at all.
• Tone in AI: the net tone of what the answers say about your subject, from −100 (all negative) to +100 (all positive), measured towards the subject rather than on the text as a whole.
• Recommendation rate: among the answers that mention you, the share in which the engine explicitly recommends you.
• Share of voice in AI answers: your mentions as a share of the mentions of you and the competitors you declared.
• Citations: the pages each answer cites, which enter the source graph as Cited pages.

Every rate is shown with its 95% confidence interval, because answers vary from one run to the next. Answers are compared engine by engine on the same prompt, so you can see, for example, that one engine recommends you in Italian while another lists only competitors in French.

What shaped it: which pages are behind the answers?

What shaped it is the Tonecast view of the source graph: the set of web pages on which what AI says about your subject was formed. You never choose these sources. Tonecast discovers them automatically, gives each one an influence weight, and records the reason it was included.

Each page is analysed like an answer: relevance to your subject, tone towards it, and the factual claims it makes. The influence weight combines how often an engine cites the page, how much that engine typically draws on that kind of source, the page's position in the search results for your prompts, and its freshness. Weights are re-estimated every week and never overwritten: the history shows how stable they are.

Every page in the graph carries its inclusion reason (for example "cited by Perplexity on 3 prompts" or "top-5 search result for a comparison prompt in French"), so nothing appears without an explanation.

Claim check: is what AI says about you correct?

Claim check compares the factual statements AI assistants make about your subject with your fact sheet, the list of verified facts you maintain (prices, features, dates, locations, what the subject is not), each with a validity date. Every claim found in an answer is classified as:

• Consistent: it matches the fact sheet.
• Incorrect: it contradicts the fact sheet.
• Outdated: it matches a value that was true in the past but is no longer valid.
• Unverifiable: the fact sheet does not cover it.

Each claim keeps the original sentence, the engine, and the cited pages that carry it, so you can see which page keeps repeating an old price. Claim check reports and documents; it never writes corrective content on your behalf. Claims are also compared across languages, so a value stated in another language is checked as the same fact.

Early signals: are the sources moving before the answers?

Early signals is the Tonecast lead indicator. It checks, for each engine, whether changes in the sources that shape its answers came before changes in the answers themselves. It watches three kinds of signal:

• the Formative Sentiment Index (FSI): the net tone of the source graph, weighted by influence weight, on the same scale as Tone in AI;
• claim prevalence: the weighted share of sources that support a given claim, such as "prices went up" or "delivery is slow";
• visibility precursors: new high-weight sources that mention you (or a competitor) and are not yet cited by the engines.

The indicator has three states. None: no lead was found over a long enough period. Observing: there is not enough data yet, or the pattern is not statistically reliable. Active: sources have moved before answers by an estimated number of days, and the pattern passed a significance test. Because Tonecast tests several signal pairs per engine, it also applies a correction for multiple testing, so that one lucky correlation among many cannot switch a signal on. Only an active signal produces an early warning. When the evidence is not there, Tonecast says so instead of guessing.

Gap & action list: what should you do next?

The gap & action list turns the findings into a list you can work through:

• prompts where competitors appear and you do not;
• incorrect or outdated claims, with the pages that carry them;
• negative formative sources with high influence weight;
• information gaps: facts from your fact sheet that no engine repeats, which are candidates for a page of your own;
• your own pages that the engines cite but that could not be read;
• active early signals, such as a claim spreading in the sources that has not reached the answers yet.

Tonecast does not generate content for you. The list tells you where the problem is and which page to look at.

Methodology card: can you defend these numbers?

Every campaign has a methodology card: the prompts and languages tracked, the engines queried, how many samples were collected per prompt and engine, the sampling window, the confidence intervals of each metric, the state of the early signal and the measured accuracy of the models behind the scores against Tonecast's published quality targets. It is written to be shown in a meeting. Read the full methodology.

Exports and reports

AI answers and the source graph can be exported as CSV, with the filters you have applied. Plans also include a PDF report (all plans) and PowerPoint export (Pro and Agency); see the table below for what is available today.

Which plan includes which feature?

Every plan includes AI Answers, What shaped it, claim check, the Early signals view and the methodology card. The plans differ in how much they track and in the features listed below. Items marked Rolling out are part of the plan and are being released during the launch period; the public demo shows what is live today.

Limits (prompts, engines, campaigns, source pages, history) are on the pricing page.

Questions about features