We Need to Support Authors Better to Deliver Accessible Content

From ATAG 2.0 to Local AI & Automated WCAG-EM

Mike Gifford

FOSDEM 2026 | Collaboration and Content Management | Open Web Alliance

Welcome everyone. Thanks to the Collaboration and Content Management track for centering this conversation on authors, not just auditors.

The Definition of Insanity

The Cycle of Failure:

We train authors → they leave → we retrain.
Authors are SMEs, not accessibility experts.
The CMS interface is the filter. If the filter allows garbage, we get garbage.

Remediation costs more than prevention every single time.

Reinforce the burn rate: every handoff and retrain cycle multiplies cost. Prevention is cheaper than cleanup.

The Missing Standard: ATAG 2.0

Part A: The Editor UI

The authoring interface itself must be accessible.

Part B: Support Accessible Content

The tool guides authors to produce accessible output.

Part B is the most ignored W3C standard. We wrote the checklist but never shipped the software because static rules could not cover dynamic context.

Stress the historical gap: we treated ATAG Part B as aspirational because we lacked contextual engines.

The We4Authors Lesson

Drupal, Joomla, Plone, Umbraco aligned on defaults like compulsory alt text.
Static code turned into Clippy: annoying, context-blind, brittle.
It could not tell a layout table from a data table.

We hit the wall of what if/then statements can do.

Call back to the pain of static linters: they either under-fire or over-fire. Authors tuned them out.

Move from Gatekeeping to Guiding.

Set the pivot: we are not scolding authors, we are giving them smart, contextual guidance.

The Moment of Authoring

Privacy-first AI pipeline:

Draft stays inside the CMS boundary.
Inference happens via WebAssembly or a local service.
No drafts leak to OpenAI or other third parties.

Local SLM stack:

Bundled with the CMS for offline and regulated deployments.
Optimized prompts for authoring context (field, template, role).
Deterministic fallbacks when the model cannot decide.

Emphasize the procurement reality: government and enterprise cannot ship drafts to remote AI. Local SLMs are our compliance path and our open-source advantage.

Practical ATAG 3.0 Moves

Contextual alt text: AI suggests descriptions using image analysis plus page context.
Plain language nudges: flag jargon and propose simpler synonyms to reduce cognitive load.
Structure hints: "You bolded this—did you mean H3?"

Goal: lower cognitive load, not replace authors.

Reiterate: keep authors in control. AI proposes, humans confirm.

Compliance is Science (WCAG-EM)

ATAG is about writing, but WCAG-EM closes the loop on managing the site.

The worst step: WCAG-EM Step 3 — select a representative sample. Humans hate doing this.

Why are we manually listing URLs for an audit in 2026?

Frame compliance as a repeatable experiment. Sampling is data science, not busywork.

The "Generate Sample" Button

The CMS already knows:

Active templates and content types.
Traffic and engagement patterns.
Editorial ownership and workflow states.

The feature: auto-generate the WCAG-EM sample and keep it current.

Turn the audit from a consulting contract into a dashboard widget.

Highlight how automation de-risks audits: consistent sampling, faster retests, cheaper renewals.

Community Update (Dec 2025)

W3C ATAG Community Group revitalized on Dec 05, 2025.
Focus: drafting Assisted Authoring guidance.
Goal: prevent AI from hallucinating accessibility metadata.

Join the GitHub discussions and shape the norms before platforms lock them in.

Invite contributors by name if present; point them to repos and meeting cadence.

The Open Web Alliance Mandate

Join the W3C ATAG Community Group.
Contribute to the shared dataset for accessibility SLMs.
Ship a WCAG-EM module in your CMS.

Challenge the room: we either build the open stack now or live under proprietary audit bots.

Resources

QR code placeholder for quick access.

Encourage scanning the QR and committing to one action before leaving the room.