How it works
The platform takes you from a live website to a ranked list of substantiated actions. Each step produces evidence the next step builds on — nothing is asserted without a basis you can inspect.
-
1
Discover
Crawl the site — from a seed URL, XML sitemaps, and in-page links — honouring robots.txt, and evaluate every page against the active rule catalogue. Deterministic coverage is 100% of crawled pages.
-
2
Understand
Open any finding to see its stored evidence: the URL, the element, the selector, the excerpt, and the HTTP context. Request a plain-language explanation grounded in that exact evidence.
-
3
Prioritise
Rank findings by a deterministic function of severity, breadth, confidence, and effort — a total ordering across the whole site, with the formula shown beside the list.
-
4
Act
Get the specific fix for the specific page: the JSON-LD to paste, the author fields that are missing, the question the content does not answer.
-
5
Monitor
Re-scan on schedule and compare any two scans — what is new, fixed, and regressed — without confusing a methodology change for a site change.
The AI boundary is structural
The language model may explain findings and draft suggested fixes. It never originates or alters a score, severity, priority, confidence, URL, or metric — those fields are simply not writable from model output. When the model is unavailable, the deterministic finding still renders in full; it is never withheld and never replaced by a canned paragraph.