AI is terrible at accessibility out of box, so I built an AI prompt pack to help power accessibility work + why “fix it in post” is now a lawsuit waiting to happen

Source: Nate’s Newsletter Date: 2025-12-06 URL: https://natesnewsletter.substack.com/p/grab-the-4-accessibility-prompts

Summary

Nate argues AI systems are fundamentally inaccessible to disabled people by default — 1.3 billion people (16% of global population) whose perspectives are absent from most training data. Early models treated disability as tragedy (excessive apologizing), reflecting training data bias rather than model intent. Beyond ethics, this is a legal risk: “fix it in post” is now a compliance liability. The piece provides four concrete audit prompts for pre-commit checks and vendor evaluation.

Implications

• Enterprise adoption thread. Accessibility failures in AI-generated content and AI-powered products create legal exposure under ADA and equivalent global frameworks. Organizations deploying AI without accessibility audits are creating compliance liabilities they may not discover until litigation.
• AI product positioning thread. The training data gap for disability perspectives is a systematic bias that requires deliberate upstream design choices, not downstream fixes. AI product teams that address this at the model interaction layer (system prompts, audit gates) will have a differentiated enterprise compliance story.
• Watch: Whether AI accessibility auditing becomes a required enterprise procurement criterion, similar to how WCAG compliance became a standard requirement for enterprise software.