Drupal AI in action: 13 AI features at a glance
This is what AI in Drupal looks like in day-to-day editorial work: from style guide checks and one-click translation to AI search with citations. Every feature is built on official modules of the Drupal AI Initiative - and can be integrated into your website on its own.
About this overview
All screenshots are taken from the public Drupal AI demo by the Drupal AI Initiative. The demo shows a fictional university, “Northmoor University”.
Descriptions and assessment written by us. We find out which features pay off for your website in the readiness check.
Editorial & quality
AI content review: every page checked against the style guide
The AI checks every page against your style guide - tone of voice, inclusive language, readability. It scores each criterion from 0 to 100, explains the score in plain words and proposes concrete rewrites.
Your benefits
- Editors see the score while writing - as a traffic light right in the editor
- Accepted suggestions become a checklist inside the edit form
- Traffic-light overview of all pages in the content list - ideal for editorial leads
- Every review is logged in an audit-proof way (score, revision, model)
- Criteria and thresholds are configurable, e.g. for SEO, legal wording or accessibility
How it works
Each content type has a review rule with several criteria. Every criterion consists of an AI agent, a scoring rubric and scored examples for calibration. Scores are graded against thresholds; scheduled re-reviews via cron are optional.
Modules
AI translation with one click
Write once, get a complete draft translation with one click - as an ordinary Drupal translation that goes through the editorial workflow. Language variety and style live in an editable prompt.
Your benefits
- Language varieties under control, e.g. Swiss or Austrian German, formal or informal, with a glossary
- HTML structure, names and numbers are preserved
- Provider-agnostic: OpenAI, Anthropic, Mistral or local models
- Cost estimate per translation right next to the button
How it works
AI Translate uses a separate prompt per target language (AI Prompts) and the model selected in the AI settings.
Modules
AI disclosure: telling readers when AI helped write
A disclosure field for all content types: readers see a card with the “AI modified” icon, how AI was used and who holds editorial responsibility - machine-readable as schema.org JSON-LD and with a report on the status under the EU AI Act.
Your benefits
- Supports implementing Art. 50 of the EU AI Act - with an automatic assessment of whether a label is mandatory
- Profiles such as “AI drafted, human reviewed”: editors pick a situation instead of filling in a form
- Every card links to your AI policy
- Filterable compliance report across all pages
How it works
A field, profiles and a site policy (always disclose or only when required). The output is a card, a meta tag and JSON-LD.
Modules
Classification & media
AI comment moderation: every comment read before it is published
Comments are checked automatically before publication: personal attacks are refused straight away, everything else is judged in the context of the post and either published or held for moderation - with a reason.
Your benefits
- Decisions by context, not keywords: the same text can fit under one post and be off-topic under another
- Hard rules such as insults apply before anything is saved
- Moderators get a short explanation for every held comment
- Visible, editable workflow (ECA) with a live test mode
- Works for any text field: webforms, user profiles, document descriptions
How it works
Three building blocks work together: a Field Validation rule as a hard rule via prompt, two AI Automators for verdict and reason, and an ECA workflow that publishes or holds. Classic regex rules can be added, for example for the correct spelling of the brand name.
Modules
Automatic tagging: tags for every page, generated from the text
The AI reads the rendered page and assigns up to five tags. Existing terms are reused, missing ones are created - on demand via a button while writing or automatically on save.
Your benefits
- Consistent topics across news, courses, programs and pages
- Foundation for related content, filters and search facets
- No more manual tagging - existing content can be tagged too
- Ordinary taxonomy: rename, merge and delete as usual
How it works
An AI Automator (“LLM: Taxonomy”) sits on a shared tag field used by all content types. The prompt and the number of tags are editable.
Modules
AI image validation: what the media library refuses
Every uploaded image is checked by a vision model before it is saved - for general image safety and against your own publication rules, for example regarding personal rights.
Your benefits
- Write your own rules in plain language - no training required
- Fine-grained: a formal reception passes, a house party does not
- Applies everywhere: form, media library in the editor, JSON:API and AI agents via MCP
- Clear error message for editors, nothing gets saved
How it works
Two Field Validation rules on the image media type, each consisting of a prompt and an error message.
Modules
AI for images: alt text, focal point, description and tags
A vision model takes over tedious image housekeeping: alt text and focal point with one click, image description and tags automatically on save.
Your benefits
- Accessibility: alt text for every image, in the language of the media item
- The focal point keeps the person in frame on social media crops
- Searchable media library via description and tags
- Manual input is never overwritten
How it works
AI Automators fill in description and tags when the fields are empty. Buttons for alt text and focal point sit right in the upload form.
Modules
Search
Ask AI: ask the website a question, get an answer with sources
Instead of keyword search, a question box: visitors ask in their own words, the website finds matching pages and an AI agent writes a short answer - exclusively from your content and with numbered citations.
Your benefits
- Every statement is backed by a source and can be verified
- No invented facts: if the information is missing, the answer says so
- Follow-up questions in the same conversation, answers in the language of the question
- Thumbs-up/down ratings reveal where content is missing
- Built on the standard Search API - fits any Drupal architecture
How it works
A vector index (Postgres with pgvector) via Search API, in which content is embedded in chunks. Plus an AI agent with a RAG tool and a listing tool, whose instructions are editable text.
Modules
Page building & agents
Canvas AI: build pages and components by chat
Two AI agents work inside the visual page builder Drupal Canvas: the Page Builder assembles a page from existing components based on a description, the Component Agent writes new React components.
Your benefits
- Plan first, then build - editors confirm before anything is built
- Uses your own components and therefore your design system
- Everything stays ordinary, editable Canvas content; nothing goes live without review
- New components with editable properties, without touching code
How it works
The Canvas AI module provides two agents in the AI panel of the Canvas editor.
Modules
Claude.ai & MCP: edit website content from a chat
The website exposes its content tools via MCP (Model Context Protocol). An AI client such as Claude.ai is connected once, signs in with the editor’s own account and can then find, read and change content - with exactly that account’s permissions.
Your benefits
- Editors work in natural language, without the Drupal interface
- OAuth sign-in, no API key handling
- Drupal permissions still apply; every change lands in the revision history
- Works with any MCP client that supports OAuth
How it works
The MCP Server module with OAuth and tool bridge submodules; the tools come from the Tool module.
Modules
Operations & control
AI observability: see every request - and what a guardrail blocked
Every AI call of the website is logged: provider, model, tokens, prompt and response. Guardrails, for example to protect personal data, stop requests before they leave the website - traceable in the report.
Your benefits
- Complete audit trail: who sent what, which rule stopped it and when
- PII guardrails for email, phone, credit card and IBAN - the data never leaves the website
- Reports by volume, model and tokens
- Prompt logging can be switched off when content must not be stored (GDPR)
How it works
The Observability and API Explorer submodules of the AI module plus guardrail sets (PII, prompt safety). The reports are Views and therefore customisable.
Modules
AI metering: token usage and costs at a glance in Drupal
For every AI call, the module records who triggered it, which model answered, how many tokens were used and what it cost - with dashboards per person and role as well as personal token budgets.
Your benefits
- Cost transparency without an external tool
- Monthly budgets and alert thresholds per person
- Editors see their own usage
- Current prices via the LiteLLM price list, CSV export, display in euros
Provider freedom: your own AI provider or your own API key
All features run through the Drupal AI module and are not tied to one provider: amazee.ai, OpenAI, Anthropic and many more can be assigned individually per capability - chat, translation, image analysis.
Your benefits
- No vendor lock-in
- A different model per capability is possible
- Keys stored securely in the Key module
- Switch providers without rebuilding features
How it works
Provider modules for the AI module, API keys via the Key module, default models in the AI settings.
Modules
Which of these features pays off for your website?
In the Drupal AI readiness check we review the version, modules, content structure and hosting of your website - and show you which use case saves you the most time. Without lock-in to a single AI provider.
Prefer to click through yourself? The Drupal AI demo is free to try.