About Tonecast

Why does Tonecast exist?

People increasingly ask AI assistants the questions they used to type into a search engine: which product to buy, whether a company is reliable, who the leading options are. The answer they read is often the only one they see. If it leaves a brand out, quotes an old price or describes a discontinued product, the brand rarely finds out.

Knowing what AI says is only half of the problem. The answers are shaped by web pages: the ones the engines cite, and the ones they typically read without citing. Without the pages, a marketing or communications team can see a number move but cannot tell what to change. Tonecast's tagline states the whole idea: Know what AI thinks. And why.

What principles does Tonecast follow?

• Precision first. The quality of the analysis (is this about the subject, what tone does it take) is measured against published targets on labelled data, not assumed.
• Radical simplicity. A campaign is set up in about five minutes by describing the subject. Every number opens the answers and pages behind it.
• The sources of the AI, not sources chosen by the user. Tonecast analyses the pages the engines cite and typically draw on, found automatically. Engines are queried only through official APIs, and sources are read through open APIs and respectful crawling.

Can the sources predict what AI will say?

Tonecast is built on a hypothesis, and treats it as one: because AI answers are formed from web pages, a change in those pages may show up before the change in the answers. For engines that search the web in real time, the path is short: new pages can be retrieved within days. For answers that rely more on what a model learned during training, the delay can be much longer.

The evidence suggests that the signal lives in the aggregate, not in any single page. One negative article does not make an answer negative. What can move answers is the weighted tone of many sources together (the Formative Sentiment Index), a claim that spreads across sources ("prices went up", "the product was discontinued"), or new high-weight pages that mention a brand before any engine cites it.

So Tonecast does not promise a forecast. It measures, for each campaign and each engine, whether the sources actually moved before the answers, and shows the result as one of three states:

• None: over a long enough period, no lead was found.
• Observing: there is not enough data yet, or the pattern is not statistically reliable.
• Active: the sources moved before the answers by an estimated number of days, and the pattern passed a significance test corrected for multiple comparisons.

If a campaign never reaches "active", the source graph still explains the answers. The value of knowing why AI says what it says does not depend on the forecast.

What Tonecast is not

• It is not a social listening tool: it does not monitor the whole social conversation, and it does not use closed platforms such as X, Instagram or TikTok.
• It is not a content generator: it shows what is wrong and where, and leaves the response to you.
• It is not a replica of anyone's chat screen: it samples the engines under controlled conditions through their official APIs and says so.

Who is Tonecast for?

Marketing managers who want to know how AI describes their brand; SEO and content teams who need to know which pages to update; PR and communications teams who track how a story reaches AI answers; and agencies who do this for several clients. See the use cases for brands, agencies, SEO and content teams and PR and communications.

How can I contact Tonecast?

Write to support@tonecast.io. To see the product first, open the public demo: it needs no account and runs on fictitious brands.