Drupal AI:
quality, grounded answers, governance
The AI demo that would have eaten the whole keynote gets its own playground instead: Northmoor University at drupal.org/ai/demo, preloaded with credits so you can just play.
Live from the stage: below, the content-review dashboard from the tutorial — animated the way it feels when an editor opens it and sees the damage, lowest score first. (Scroll it into view.)
Tutorial: Mastering Drupal AI — Content Quality, Grounded Search, Assistant Integration & Governance
Based on the DrupalCon Keynote Demonstration featuring Northmoor University (drupal.org/ai/demo).
1. Overview: The Problem Drupal AI Solves
Large-scale websites face a common set of content management challenges:
View source diagram
+-----------------------------------------------------------------------+
| THE PROBLEM |
| |
| [ Programs ] [ Courses ] [ People ] [ News ] [ Events ] |
| [][][][][] [][][][][] [][][][][] [][][][][] [][][][][] |
| [][][][][] [][][][][] [][][][][] [][][][][] [][][][][] |
| |
| * THOUSANDS OF PAGES |
| * MANY EDITORS |
| * POOR QUALITY CONTENT IS HARD TO SPOT |
+-----------------------------------------------------------------------+Every page on an organization’s website represents a promise to visitors. When information is outdated, inaccurate, or poorly written, it becomes a business risk. Drupal AI introduces automated review, content auditing, conversational retrieval, and external AI connectivity directly within the CMS.
2. Testbed Architecture: Northmoor University Demo
To demonstrate real-world applicability on complex web architectures, Drupal AI provides a full-scale reference site:
- Demo Environment: Northmoor University
- Live Demo URL:
drupal.org/ai/demo - Content Taxonomies Demonstrated:
- Programs
- Courses
- People (Faculty & Staff)
- News
- Events
- Policies
3. Module 1: Find Weak Content and Fix It
3.1 Establishing Automated Content Review Standards
Drupal AI audits every page against configured editorial criteria:
View source diagram
+------------------------------------------------------------+
| CONTENT REVIEW STANDARDS |
| |
| [ Accuracy ] [ Tone ] [ Inclusive language] |
+------------------------------------------------------------+- Accuracy: Flags outdated or contradictory statements.
- Tone: Identifies voice and style mismatches.
- Inclusive Language: Detects non-inclusive terminology.
3.2 Content Review Dashboard
Pages are evaluated, scored, and sorted so editors can prioritize problematic content:
View source diagram
+----------------------------------------------------------------------+
| Content review 32 pages, lowest first|
|----------------------------------------------------------------------|
| [ ] Marine Biology BSc [==== ] Score: 14 |
| [ ] Undergraduate Admissions [====== ] Score: 22 |
| [ ] Faculty Directory [======= ] Score: 25 |
| [ ] Campus Safety Policies [========= ] Score: 34 |
| [ ] Research Grants Overview [============ ] Score: 52 |
| [ ] Alumni Relations [====================] Score: 95 |
+----------------------------------------------------------------------+3.3 In-Context Content Remediation Workflow
When editing a flagged node (e.g., Marine Biology BSc), feedback and suggested rewrites appear directly alongside the content:
View source diagram
+----------------------------------------------------------------------+
| Content review / Marine Biology BSc |
|----------------------------------------------------------------------|
| Fieldwork begins in the first semester along coastal intertidal zones.|
| =================== [Highlighted Flagged Text] ==================== |
| |
| +--------------------------------------------+ |
| | Tone | |
| | Suggested rewrite: | |
| | "Fieldwork from the first semester, on a | |
| | shoreline that changes every tide." | |
| | | |
| | [ Accept ] [ Skip ] | |
| +--------------------------------------------+ |
+----------------------------------------------------------------------+Editor Action Workflow:
- Click
Acceptto apply the generated replacement directly into the active field. - The UI logs progress:
Diagram [✔] Accepted Applied to the page Progress: [=====> ] 1 of 2 fixed - Click
Skipto keep the existing manual copy. - Non-technical staff can maintain content without relying on specialized copywriters.
4. Module 2: Grounded Answers (RAG Search)
4.1 Retrieval-Augmented Generation on Your Content
Drupal AI answers user queries using your own indexed site data rather than generic external models. Every answer provides source citations.
View source diagram
+-----------------------------------------------------------------------+
| Ask westbrook.edu |
|-----------------------------------------------------------------------|
| > User: How do I study marine biology? |
| |
| > Drupal AI: |
| Undergraduate studies in Marine Biology require completion of core |
| coursework and field research at coastal research stations... |
| |
| Sources: |
| [ 1 Marine Biology BSc ] |
| [ 2 Entry requirements ] |
| [ 3 Field stations ] |
+-----------------------------------------------------------------------+4.2 Handling Inquiries Outside Site Scope
When a question falls outside the site's scope, Drupal AI does not guess or hallucinate answers. Instead, it directs the visitor to the relevant department:
View source diagram
+-----------------------------------------------------------------------+
| > User: How much does it cost? |
| |
| [ Outside site scope ] |
| |
| Please contact our admissions department for current fee schedules: |
| |
| [ Admissions team -> ] |
+-----------------------------------------------------------------------+5. Module 3: Connecting External AI Tools (Claude & ChatGPT)
5.1 Multi-Page Batch Updates via Assistant
Drupal AI integrates with external assistants (such as Claude or ChatGPT) to perform batch operations across site nodes:
View source diagram
+-----------------------------------------------------------------------+
| Your AI assistant (Claude / ChatGPT) |
|-----------------------------------------------------------------------|
| Prompt: |
| "Professor Vargas was promoted. Update every page that mentions her." |
| |
| Status: |
| Searching westbrook.edu ... |
| Found on 1 page... Found on 4 pages... Found on 10 pages. |
| |
| Result: |
| Found 10 pages. Drafting updates. |
| |
| [✓] Faculty Directory - Biology [ Draft ] |
| [✓] Marine Biology BSc Program Overview [ Draft ] |
| [✓] Coastal Ecosystem Research Team [ Draft ] |
| [✓] Academic Senate Board Members [ Draft ] |
| |
| [ Review 10 drafts ] |
| Nothing published. |
+-----------------------------------------------------------------------+5.2 Editorial Control: "Drafts, Not Surprises"
- Changes are generated in a
Draftstate. - Automated actions do not publish directly to production.
- Editors inspect, approve, or reject changes on a per-page basis.
- No vendor lock-in: Connect to your preferred AI model provider while maintaining control of your Drupal content repository.
6. Module 4: Enterprise Governance & Accountability
When deploying AI to public audiences, organizations need verifiable controls over output accuracy and operations.
View source diagram
+--------------------------+
| westbrook.edu |
| [ Public Visitors ] |
+-------------+------------+
|
+---------------------+---------------------+
| | |
v v v
+--------------+ +-------------------+ +------------------+
| SHIELD | | KNOWLEDGE HUB | | AUDIT TRAIL |
| AI Guardrails| |One Source of Truth| |Every Interaction |
| | | | | Logged |
+--------------+ +-------------------+ +------------------+6.1 Governance Pillars:
- AI Guardrails: Constrains model responses within defined operational and factual boundaries.
- One Source of Truth: Connects all AI tools (chatbots, page reviewers, assistant connectors) to a single, centralized Drupal knowledge repository.
- Every Interaction Logged: Records inputs, retrieved context, and generated responses to provide a complete audit trail for compliance and review.
7. Summary & Resources
- Deployment Model: Open source.
- Core Principle: "Own your Stack. Own your data."
- Collaborators: Developed by Drupal AI ecosystem partners.
- Hands-on Demonstration: Try the interactive university showcase at:
Diagram drupal.org/ai/demo
“It’s very cool. I got a demo of the demo website and I was very impressed — I recommend you all check it out. It’s great for learning what’s going on with the initiative. It can be overwhelming.”
“And because it’s completely preconfigured, you can actually use it to demo Drupal AI to your customers as well. It’s not only good for you — it’s good to win new business.”
“Just yesterday I learned that in the last several months, over a hundred people have contributed to the Drupal AI initiative. There’s a lot of momentum — there will be a blog post with more statistics and data today as well.”
“All of this is made possible thanks to the Drupal AI partners you can see on the screen. If you want to join and become a partner, it’s not too late — go talk to Dominic or fill out that form.”
“Just like the city of Rotterdam, we’ve been doing lots of scrubbing and not a lot of blabbing. All three of these things have the same goal: grow Drupal by bringing Drupal’s strengths to more people — us going to where the people are. And we’re making really great progress.”
But where are the people, increasingly? Inside AI assistants. “First, what I want to do is give you a very quick demo to show you what I mean,” Dries says — and queues up Tutorial V.