AI-generated policies are now grounded in your actual site facts

Released

Marina Köhler

This is the change that makes the AI generator stop sounding generic. Until now the generator drafted policies from a strong template plus your sector. Good output, but not specific. The new version grounds every draft in the real facts extracted from your site: the forms you have, the contact info you publish, the integrations you use, the kinds of data you collect.

The five-step pipeline now does:

Sitemap scan to discover your pages.
Content extraction to pull the meaningful prose, forms, embed iframes and contact details.
Sector detection to classify your industry (healthcare, finance, ecommerce, b2b SaaS, etc.).
LLM drafting grounded in both the sector and the extracted site facts.
Persist as draft so you can edit before publishing.

The result reads like a draft written by someone who actually looked at your website, not one that pattern-matched on your industry. We also fixed a navigation issue and a refresh edge case that occasionally lost state during long generation runs.

April 29, 2026

Activity feed

NA
Noor Abdullah
How does the generator handle a sitemap that hides behind basic auth or only exposes 2-3 public pages? Does the grounding still produce something useful or does it fall back to generic boilerplate?
May 18, 2026
•
Replay
Alex Rivera
The first draft for our project mentioned data we do not collect. Trust issue when the AI hallucinates fields. Hoping the grounding pass meaningfully reduces that class of error.
May 10, 2026
•
Replay
Yuki Nakamura
The five-step pipeline is well-decomposed. The sector classifier in particular is a smart abstraction — sector-specific clauses are exactly the part most generic generators get wrong.
May 9, 2026
•
Replay
NA
Nia Adeyemi
This is the upgrade that made the AI generator actually useful for us. The previous version produced generic policies and we had to redline heavily. The grounded version reads like it actually visited our site.
May 3, 2026
•
Replay