Modern HTML often arrives in stages
A framework application can return a small server shell, then request data and construct the meaningful DOM in the browser. Fetching initial source may therefore miss product cards, headings, metadata widgets, or content loaded after hydration.
Rendered HTML captures the DOM at a selected moment. That moment must be defined with the same care as a screenshot: viewport, readiness, target data, and cleanup can all influence the result.
Parse without trusting
HTML from an external page is untrusted input. Use a parser appropriate to the application, limit resource use, and sanitize any content that may later enter an interface. Do not execute scripts from the returned document inside a privileged environment.
Selectors should be tested against representative templates. When a target is controlled by another team or publisher, structural changes can break extraction even if the page remains visually healthy.
Preserve the source of every field
Store the requested target, observed time, rendering configuration, raw or retained HTML policy, parser version, and validation outcome with derived records. This provenance helps resolve disputes when a field changes or disappears.
Respect target authorization, robots and publisher policy, data protection, and collection limits. A rendering API does not grant permission to process content.
Use AI DOM analysis when classification or review supports a rendered DOM workflow, and AI content extraction when the documented analysis response already matches the need.
