A website brand scan reads a small, same-domain sample of your public site to stage voice guidance, design tokens, and candidate factual claims. It runs only after you explicitly accept the scan and provide the URL; Deloosh never starts it unprompted.
What the scan reads
The scanner fetches the root page and may follow up to two additional links from the same domain, for no more than three pages in one run. It records the visited URLs and source hashes with the result.
For every factual claim it extracts, the scanner requires a source excerpt that appears byte for byte in the fetched page. A claim pointing to an unfetched page or a non-verbatim excerpt causes the scan to fail instead of creating weak evidence.
What happens after extraction
Factual claims are persisted as candidate, never as trusted proof. Voice and design output are staged with the scan result. The backend promotion route exists, but no Workbench or assistant control exposes it to customers today. A completed scan by itself does not promote staged output or authorize draft copy.
The current plan configuration lists one included brand scan and a 30-day rescan interval for every plan. Those are shipped plan settings, not a promise that the numbers will never change.
For claim review, see BRAND-2, Confirming or rejecting scanned claims. For the broader control boundary, see TRUST-1, What Deloosh does on its own — and what it always asks first.