A public site does not need an AI call when a reader opens an article. Content can be drafted, reviewed and stored in versioned JSON, leaving the site predictable and available during external outages.

The useful answer depends on the purpose of the page, the people allowed to edit it and the cost of an error. A reliable workflow separates editorial decisions, technical delivery and final verification instead of hiding all three inside one automated operation.

The decision to make first

Generation belongs to production, not public rendering. The final record stores title, slug, reviewed HTML, category and dates for pages, sitemap and RSS.

A practical method

  1. Define a stable JSON schema and required fields.
  2. Draft and verify outside the public request path.
  3. Validate slugs, links, dates and allowed HTML.
  4. Deploy JSON with the code that reads it.
  5. Keep a backup and test rollback.

Checks before publishing

  • The source of truth is explicit and can be restored.
  • URLs, dates and visible claims match the delivered page.
  • A human reviewer can reproduce the check without a hidden service.

The main pitfall

One missing comma can invalidate JSON, so every release must decode the complete file before publication.

Recommended next step

Add JSON-load and HTML-render tests, then keep every editorial change in Git.