https://events.drupal.org/rotterdam2026/session/ai-augmented-accessibility-automated-alt-text-governance-gates

Stop reacting to accessibility errors and start automating inclusive governance. Learn how to use AI to find accessibility errors earlier and fix them faster. We’ll bridge the gaps between AI, code, and humans with clear guidance for everyone.

Prerequisite

It would be useful to be familiar with some basics of accessibility. This is not a 101.

Target Audience

Anyone working to meet accessibility targets.

Outline

Accessibility is often the last gate in a project. Like many bugs the longer they wait, the more difficult they are to fix. Learn how we can "shift left" by using AI not just to find errors, but to build better systems. This session provides a practical roadmap for integrating AI into the full Drupal lifecycle:

* Content: Automating descriptive alt-text & transcripts using Drupal AI Core.

* Infrastructure: Using LLMs to architect & configure CI/CD accessibility gates (Axe-core/Playwright).

* Triage: Leveraging GitHub’s AI-powered accessibility toos to turn scans into Pull Requests.

* Governance: Implementing ACCESSIBILITY.md or Claude skills to produce better results.

* Hierarchy of Truth: Adding AI roles to that of other automated tools & humans—why the user remains the ultimate source of truth.

Join us to move beyond "compliance checklists" & toward a scalable, AI-assisted inclusive web.

Mike will also be tying this into the book he recently co-authored. Digital Accessibility Ethics: Disability Inclusion in All Things Tech,

Learning Objectives

* Automate Content Workflows: Set up the Drupal AI module for automated, human-verified media accessibility.

* Architect Testing Gates: Use AI assistants to write configuration for robust CI/CD frameworks.

* Implement Modern Standards: Learn to use ACCESSIBILITY.md & Claude Skills to keep projects transparent and compliant.

* Validate Strategically: Understand the "Hierarchy of Truth" to know when to trust what.

• CI/CD - Continuous Integration / Continuous DevelopmentMoving from reactive post-launch fixes to continuous pipeline-level governance.

Automated site scanning tool.

What folks want.

• Rule 1: Don't use a LLM where a simpler algorithm will do.
• Rule 2: Trust people over machines

1. Accessibility Statements, Users with Disabilities
2. Professionals
3. Standard automated tools
4. AI driven tools
5. AI driven answers

Culture Dictates What Is Acceptable: Compliance is an organizational.

Repo-Level Guardrails (ACCESSIBILITY.md):

• Commit standards directly into project roots alongside README.md.
• Provide machine-readable instructions that both human developers and coding LLMs must obey.
• accessibility-skills too

• Alt text can be rated with an algorithm
• A LLM can probably do it better than a person.

• Scan a site that has already been scanned and demonstrate the LLM Prompt
• Best experience is still with data and someone with User Researcher

• Note this hasn't been tested to see if it works

• Note this hasn't been tested to see if it works

• The W3C is working on this.

• I am still working on this.

• Sample URLs:
  • https://mgifford.github.io/open-scans/reports/issues/issue-353/2026-09-14T14-12-20-749Z/report.csv
  • https://mgifford.github.io/open-scans/reports/issues/issue-353/2026-09-14T14-12-20-749Z/report.csv
  • https://mgifford.github.io/open-scans/reports/issues/issue-350/2026-09-14T13-56-56-213Z/report.csv
  • https://mgifford.github.io/open-scans/reports/issues/issue-351/2026-09-14T14-44-51-809Z/report.csv