Dyslexia affects how roughly one in ten people decode written language. AI tools can simplify syntax, increase line spacing, and produce text-to-speech audio in minutes. Those outputs help only when they follow evidence-based formatting principles from organizations like the British Dyslexia Association and WCAG readability guidance, not when they replace editorial judgment with generic "simplify" prompts. Teams building AI writing pipelines or pairing written content with AI voice narration need a workflow that treats accessibility as a design constraint, not a post-publish checkbox.
Accessibility Principles That Matter for Dyslexic Readers
Dyslexia-friendly content reduces visual crowding, shortens sentence load, and offers multiple access paths (visual text plus audio) without changing factual meaning. Research-backed recommendations include generous line spacing (around 1.5 times or more), left-aligned text, short paragraphs, clear headings, and avoidance of justified columns that create uneven word gaps. Fonts marketed for dyslexia, such as OpenDyslexic, remain debated: some readers prefer them; others find standard sans-serif faces like Arial or Verdana equally effective when spacing and structure are strong.
WCAG 2.2 does not mandate a dyslexia-specific font, but Success Criterion 1.4.12 (Text Spacing) requires that users can adjust spacing without loss of content or function. That means your CMS, PDF exports, and email templates must survive increased line height, letter spacing, and paragraph spacing. British Dyslexia Association guidance emphasizes cream or off-white backgrounds over pure white glare, sufficient contrast without harsh black-on-white extremes, and bullet lists instead of dense prose blocks.
| Principle | Research-backed practice | Common AI mistake |
|---|---|---|
| Spacing | 1.5+ line height, short paragraphs | Rewrites text but leaves cramped CSS |
| Sentence length | One idea per sentence where possible | Choppy fragments that lose meaning |
| Structure | Headings, bullets, predictable layout | Wall of simplified prose |
| Multimodal access | TTS audio aligned to same text | Audio that drifts from on-page copy |
AI Formatting Tools in the Workflow
Stage one: draft or import source content. Stage two: run AI passes for syntax simplification, glossary insertion, and heading structure. Stage three: apply layout rules in the publishing system, not only in the AI output pane. Large language models in AI writing assistants can rewrite passive voice, split long sentences, and define jargon inline. Dedicated readability plugins and CMS extensions automate some spacing and font defaults. Text-to-speech tools from AI voice vendors generate narration tracks for articles, onboarding docs, and course modules.
Prompt Patterns That Work
Instruct the model to preserve technical accuracy, keep proper nouns unchanged, and limit sentence length to roughly twenty words unless a longer sentence is required for legal precision. Ask for bullet summaries after each major section. Request a plain-language glossary table for acronyms. Avoid prompts that say "dumb it down," which often remove qualifiers readers need for informed decisions.
TTS and Dual-Format Publishing
Generate audio from the final approved text, not from an early draft. Sync paragraph breaks in the audio player with on-page anchors so readers can switch between listening and scanning. ElevenLabs, Amazon Polly, and platform-native TTS differ in pronunciation of product names; maintain a custom pronunciation lexicon for brand terms and medical vocabulary.
Readability Metrics and When to Trust Them
Flesch-Kincaid, SMOG, and automated readability scores estimate syntactic complexity. They do not measure dyslexia accessibility on their own. A paragraph can score "easy" while using visually similar words, irregular spellings, or idioms that trip decoding. Use metrics as guardrails: flag sections above a grade level target for your audience, then apply human review. Pair scores with spacing checks and a quick TTS listen-through, which often surfaces awkward rhythm that formulas miss.
| Metric | What it measures | Limitation for dyslexia |
|---|---|---|
| Flesch Reading Ease | Sentence length and syllable count | Ignores layout and font choices |
| SMOG index | Polysyllabic word density | Penalizes necessary technical terms |
| Automated WCAG scanners | Contrast, structure, ARIA | Rarely judge cognitive load of prose |
Human Review Checklist
Every AI-assisted dyslexia-friendly pass needs an editor or subject-matter reviewer who checks meaning, tone, and inclusive language before publish. Reviewers should confirm that simplification did not drop warnings, dates, or conditional phrasing ("may," "unless," "in some cases"). Run a quick read-aloud test: if the reviewer stumbles often, dyslexic readers likely will too. Where possible, include feedback from readers with dyslexia in user testing; no proxy checklist replaces lived experience.
- Compare AI output against source for factual drift.
- Verify headings follow a logical outline without skipped levels.
- Confirm links and button labels make sense out of context.
- Listen to TTS for mispronunciations and pacing issues.
- Test page with user-controlled text spacing per WCAG 1.4.12.
Implementing in CMS and Design Systems
Accessibility improvements fail at publish time when templates hard-code tight line-height or disable user zoom. Map dyslexia-friendly defaults into your design system: body line-height 1.6 or higher, paragraph spacing of at least 1em, max line length around 70 characters for long articles, and focus styles that remain visible on cream backgrounds. WordPress, Contentful, and headless CMS setups should expose a "readable article" content type with these tokens baked in rather than asking editors to paste inline styles per post.
When AI rewrites content inside a CMS plugin, store both the canonical source and the accessible variant. Legal and medical teams may need the precise original phrasing for audit trails while the public page shows the simplified version. Version diff tools help reviewers see what the model changed. Schedule periodic re-runs only when source facts change, not on every typo fix, to avoid drift.
Email and PDF Edge Cases
Marketing email clients strip custom fonts and ignore some spacing CSS. Test dyslexia-friendly newsletters in Gmail, Apple Mail, and Outlook with increased text size enabled. For PDFs, export from structured HTML or Word with tagged headings rather than printing web pages to PDF, which flattens spacing controls. AI can rewrite PDF copy in source documents, but automated PDF remediation tools remain imperfect for complex layouts.
Font and Layout Limits
Do not rely on a single "dyslexia font" toggle as your entire accessibility strategy. OpenDyslexic and similar typefaces may help some users; others report no benefit compared with well-spaced Arial. British Dyslexia Association materials note that font choice is personal and that spacing, structure, and background color often matter more. Minimum body text around 12pt print equivalent (roughly 16px on web) is a practical floor; many teams target 18px for long-form education content.
Avoid all-caps blocks, italic-heavy paragraphs, and text overlaid on busy images. If you must use columns, keep them for short callouts only; multi-column article bodies increase return-sweep errors for dyslexic readers. Dark mode themes need the same contrast discipline as light mode; washed-out gray body text fails WCAG even when it looks modern.
Measuring Success With Readers
Track completion rate, time-on-page, support tickets about comprehension, and optional reader surveys rather than readability score alone. A lower Flesch score means little if task completion improves for dyslexic users in moderated usability sessions. Partner with disability employee resource groups or student accessibility offices when testing internal documentation. Publish a short accessibility statement describing spacing controls, TTS availability, and how to request alternative formats.
Combine quantitative metrics with qualitative feedback forms that ask whether spacing, audio, and sentence length helped or hindered. Iterate prompts and CSS tokens quarterly rather than treating accessibility as a one-time AI rewrite project.
Frequently Asked Questions
Can AI fully replace accessibility editors?
No. AI accelerates simplification and audio production, but it cannot certify that content meets organizational reading level goals, legal accuracy, or individual reader needs. Keep human review in the workflow.
Is OpenDyslexic required for dyslexia-friendly sites?
No major standard requires OpenDyslexic. Offer readable defaults (sans-serif, adequate size, spacing) and respect user agent font overrides. Some teams provide an optional dyslexia-friendly theme without forcing it globally.
Does TTS help all dyslexic readers?
Many dyslexic readers benefit from listening while following along; others prefer visual text with spacing adjustments only. Provide both without making audio mandatory for access to information.
What reading grade level should we target?
Public health and citizen-facing content often targets grade 6 to 8 in U.S. terms, but technical and professional audiences may need higher complexity with strong structure. Match level to audience and document the choice.
How does this relate to WCAG compliance?
Dyslexia-friendly formatting supports several WCAG goals (readable text, spacing, predictable structure) but WCAG conformance does not automatically mean optimal dyslexia support. Treat WCAG as a baseline and dyslexia guidance as additive UX research.
Should we use AI to fix PDFs?
AI can rewrite PDF source copy, but tagged PDF structure, reading order, and export settings still require proper authoring tools. A simplified Word document exported as an untagged PDF remains difficult for screen readers and reflow tools.
Which AI writing tools fit this workflow?
General-purpose LLMs, grammar assistants, and CMS-specific readability plugins all play a role. Choose tools that allow custom style guides, export clean HTML, and integrate with your AI writing approval chain rather than one-click publish from the model interface.
Should students get different formats?
Educational accommodations may require large print, audio, or simplified summaries alongside standard texts. AI can produce variants in parallel when teachers validate content equivalence for assessments.
