AI transparency

Version 2026-10-19 draft · The legal documents are published in English. The English version governs.

What AI does

Four tasks, each with a fixed prompt, a closed output format and a code check on the answer:

What AI never does

Which data it sees

Excerpts of your site's pages and the evidence behind an issue. Data from Google Search Console reaches a model only after an owner or admin of your organisation records consent in Settings, and every use is written to your audit log. Content read from other sites is passed as untrusted data and labelled so.

Which models

Requests go through a gateway we operate to the vendors on the sub-processors page, addressed by capability tier rather than by vendor, so the vendor behind a task can change without changing the product. No vendor may train on your data. Enterprise organisations can require zero-data-retention routes and a region.

How it is paid for

Each task has a published credit price. Credits are charged before the call and refunded if the model gives no valid answer. Your plan's monthly credits reset each month; purchased top-ups do not expire while you subscribe. A daily spending ceiling per organisation protects you from runaway use.