This morning I attended "AI, A11y, and WordPress: Deciding What to Trust, Test, and Do Yourself" by Simon Miner at WordPress Accessibility Day 2026.
One line from the session stayed with me:
"AI can write the patch, but I still have to know what 'fixed' means."
My key takeaways:
🔹 AI-generated code is not accessible by default. Simon pointed out that this year's WebAIM Million report showed a regression for the first time, at the same time that AI-assisted coding is growing fast.
🔹 Use AI for specific, bounded, reviewable fixes, like keyboard controls, semantic markup/ARIA and focus behaviour. The human still defines the problem, reviews the code and tests it.
🔹 Only rely on AI for work you can do and review yourself. You cannot accurately review what you don't understand.
🔹 Better alt text prompts: tell the AI what the image is OF, what it is FOR, and what the content is ABOUT.
🔹 Screen reader testing still catches issues that automated scans miss. Code and visuals can look correct and still fail for assistive technology users.
My biggest takeaway: in an AI-driven world, accessibility expertise matters more, not less.
Thank you, Simon Miner and the #WPAD2026 organisers, for a practical and honest session.
Slides: https://lnkd.in/gCBT3aHF
From lnkd.in