Drupal has always been good at one thing: structured, governed, trustworthy content. AI search should build on that. It shouldn’t work around it.
This session shows Drupal acting as the content source it was always meant to be. An AI agent on the Drupal site answers visitors’ questions in plain language. Every answer is built only from the site’s own content, with citations that link back to the real pages.
The key idea is that the agent doesn’t answer every question the same way. It chooses a tool for each question:
- RAG search for “how do I…” and “what is…” questions, where semantic similarity works well.
- Non-RAG listing tools for “which…” and “how many…” questions, where vector search only returns the top few matches and gives an incomplete answer. Entity listings from tools such as Tool Belt return the complete set, still fully cited.
- AI PageIndex for long, structured documents like PDFs, policies and manuals. An LLM reasons over a document tree instead of embeddings, needs no vector database, and cites the exact section.
Using the right tool for each question gives more accurate answers, and it can lower the cost per query, because not every question has to go through embeddings and vector search.
Not all content starts out in Drupal, though. We’ll use CRWLR to crawl an external website into ordinary Drupal nodes. Once there, those nodes are indexed, governed and cited like any other content.
Finally, we’ll premiere something new: a proof of concept for embedding AI Answers on any web page, whether WordPress, a static site or any other stack. You paste a small snippet into the page. Shadow DOM keeps the widget’s styling safe from the host site’s CSS. Drupal stays the source of truth, and the agent and its tools still decide where each answer comes from. Drupal becomes the answer engine behind every site you run.
You’ll leave with:
- A mental model for tool-based, cited answers rather than “just add RAG”
- When to use RAG, listing tools or vectorless retrieval, and the cost trade-offs of each
- How to bring external content into Drupal so it can be cited
- A first look at serving Drupal-grounded answers outside Drupal