Building an AI content import module for Drupal

AI can suggest content. But who decides what goes into Drupal?

Asking AI to improve a paragraph is easy. Updating a real website is a different problem. A Drupal site contains translated pages, products, services and projects, all connected through fields, taxonomy terms and references. A useful suggestion must fit that structure without accidentally changing something else.

For Visual Shapers, I am building a Drupal 11 module called AI Content Exchange. Its purpose is not to let AI take over the website, but to make the exchange between AI-assisted editing and managed content more precise and reviewable.

Exporting context, not just text

The module produces a structured JSON export of published content belonging to the selected website. Alongside editable text, it describes language versions, content types, SEO fields, taxonomy assignments and existing relationships. Protected information, including URLs and media references, is included as context rather than as permission to change it.

That distinction matters for a bilingual site with a headless Drupal backend and a static Astro frontend. English and Dutch need their own content, while some relationships are deliberately shared by both languages.

A review before any import

An AI-generated proposal is still only a proposal. The import review in Drupal compares the proposed changes with the current content. Only explicitly permitted fields may be changed. Source hashes help detect content that has changed since export, and existing URL aliases are protected.

References deserve particular care: replacing a list of selected projects or technology terms must not silently remove entries that should remain. The importer therefore works with complete, ordered lists that an administrator can inspect before approving a change.

Testing the next step: creating content and linking it

Earlier tests covered controlled edits to existing text and references. The next version extends the workflow to creating new Drupal content with both English and Dutch translations, creating relevant taxonomy terms and linking the new item to an existing page within one reviewed import.

This work story is itself the test case: create a bilingual project, assign its technologies and add it to the Selected work section on the homepage. That combined workflow is still being validated on the Drupal test environment; it should not be described as a proven production feature yet.

Human editorial control remains the point

The design principle is simple: AI can help prepare structured content, but a human reviews and authorizes what Drupal actually stores. Publishing the static Astro website is a separate, deliberate step.

The project brings together Drupal development, JSON data exchange, multilingual content modelling and careful change management. The goal is a repeatable editorial workflow, not a one-click promise of autonomous publishing.