Fix Your PDFs: Practical Accessibility Remediation That Scales With AI and Good Habits

Most teams publish documents that look fine on screen but break down the moment someone tries to use a screen reader, zoom to 200%, or navigate by keyboard. Accessibility issues aren’t just compliance bugs—they quietly lock people out of your content and cost you trust, time, and reach. The good news: with a few durable habits, purposeful remediation, and a safe dose of AI, you can turn fragile files into usable, testable, and maintainable documents.

This guide shows you how to fix existing PDFs and slides, set up authoring workflows that avoid rework, use AI for alt text and structure suggestions without losing control, and test the result with the same tools your audience uses. It’s hands-on, tool-agnostic, and designed for small teams that need reliable results—not a complicated standard to memorize.

What “accessible” means for documents

Accessibility is not a checkbox. It’s a property you can measure and improve. For documents, two anchors help keep you honest:

WCAG (Web Content Accessibility Guidelines): four principles—perceivable, operable, understandable, robust—apply to HTML and also guide PDF and EPUB content.
PDF/UA (ISO 14289): the standard for tagged, accessible PDFs. It defines how structure, reading order, alt text, and semantics must be expressed so assistive tech can work reliably.

Practically, accessible documents let people:

Navigate by headings, bookmarks, and links
Use a screen reader to read paragraphs, lists, tables, and figures in the intended order
Fill forms with clear labels and a sensible tab order
Read at high zoom or in reflow mode on small screens
Understand content regardless of color alone, with enough contrast
Switch language when words change (e.g., English paragraph with a French quote)

If your documents don’t support these, they’re excluding someone. The rest of this article shows you how to change that quickly and sustainably.

Build accessibility into authoring, not just remediation

Fixing PDFs after the fact is possible, but slower than building it in. Start with the authoring tools your team already uses and set a baseline.

Start with styles that map to structure

Headings: Use built-in Heading 1–6 styles in Word/Google Docs. Avoid “bold + bigger text” fakes. In InDesign, map paragraph styles to PDF tags (H1, H2, P) before export.
Lists: Use real bulleted/numbered list features; don’t paste symbols. In InDesign, map list styles so PDF export creates L, LI, Lbl, and LBody tags.
Tables: Use the table tool, set header rows, and mark the scope (row/column). Keep tables simple—no merged cells unless necessary. Provide a brief table summary.
Links: Write meaningful link text (“Download the Budget 2026 report”) rather than “click here.”
Language: Set the document language and mark inline language changes for quotes or terms.
Alt text: Add concise descriptions where images carry information. Mark decorative images as decorative (so they’re skipped).
Color & contrast: Ensure color isn’t the only signal (e.g., add labels/patterns) and meet contrast targets (WCAG AA: 4.5:1 normal text, 3:1 large text).

Slides and reading order

Give each slide a unique title.
Check the reading order panel (PowerPoint) or arrange layers in a logical sequence.
Don’t paste text as images; when unavoidable, provide an on-slide text equivalent.

Charts, math, and complex visuals

Charts: Include data tables nearby or in an appendix. Use alt text for the message (“Revenue grew 12% YoY, strongest in Q4”) and a longer description where needed.
Math: Prefer authoring tools that can export MathML (e.g., MathType in Word). If delivering as PDF only, provide a textual equivalent or consider an HTML/EPUB companion for math-heavy documents.

Templates that bake in success

Create and share templates with correct tag mappings, heading levels, color palettes, and alt text guidance baked in.
Include a “How to make this accessible” first page with a checklist. Authors see it while drafting, not as an afterthought.

How to fix existing PDFs—fast and thoroughly

When all you have is a PDF (or a scan), remediation is your path. The goal is a tagged PDF with a correct structure tree and a trustworthy reading order.

Step 1: If scanned, OCR with care

Use high-quality OCR that preserves layout and recognizes language correctly.
Deskew pages and remove artifacts that confuse structure detection.
If the scan quality is poor, consider re-creating from the source or offering an HTML alternative.

Step 2: Establish the document’s skeleton

Set Document Properties: Title, Author (org), Language, and set “Display Title” so readers see the title, not the filename.
Generate a structure tree via auto-tagging as a starting point, then fix it manually: ensure a single H1, descending headings (no skipping H2 to H4), and paragraphs as P tags.
Mark headers/footers and purely visual separators as Artifacts so they won’t be read aloud as content.

Step 3: Repair reading order

In multi-column layouts, use the Reading Order or Order panel to sequence content left-to-right, top-to-bottom.
For sidebars or callouts, place them logically after the paragraph that references them.
Create Bookmarks matching the heading structure to support quick navigation.

Step 4: Get lists, tables, and figures right

Lists: The structure should be L > LI > Lbl + LBody for each item. Fix any pseudo-lists auto-tagging created.
Tables: Identify header cells (TH) and set scope to row/column. Avoid complex spanning headers unless necessary, and provide a summary.
Figures: Provide alt text that fits the context. Keep it concise (often under 125 characters). If the surrounding text fully explains the figure, mark it decorative.

Step 5: Links, forms, and language

Links: Ensure link text is descriptive. If the PDF shows full URLs, add readable link text and keep the URL as the actual destination.
Forms: Each field needs a label (often the Tooltip in Acrobat), a sensible tab order, and a clear submit/reset flow. Don’t rely on color alone to show required fields.
Language: Mark spans for words in another language so screen readers switch voices correctly.

Step 6: Contrast, reflow, and final checks

Verify contrast for text and essential graphics using a trusted checker.
Test Reflow mode and scaling at 200%+ to ensure content remains readable without horizontal scrolling.
Run an automated checker as a lint pass, then do a manual screen reader test (see below).

Bring AI into the loop—safely and usefully

AI can make remediation faster, but only when you set guardrails and keep a human in charge. Use it to propose, not to publish.

Where AI helps

Alt text drafting: Provide page context and figure captions; ask the model for 1–2 concise options. Accept or edit.
Link text suggestions: Replace “click here” with purpose-driven text based on nearby headings and sentences.
Table header detection: Identify header rows/columns and propose scope; you confirm.
Reading order hints: Suggest logical sequences for multi-column pages; you set the final order.
Plain-language summaries: Generate a short abstract for the document’s first page.

Guardrails that keep you in control

Process locally for sensitive drafts. Use an on-device model runner (e.g., Ollama) when NDAs or compliance demand it.
Redact before upload if using a cloud model. Never send personal data, keys, or unreleased financials.
Style guide for alt text: Decide voice, length, and do/don’t examples. Keep a glossary for names and acronyms the model must respect.
Structured prompts: Give role (“You draft alt text for accessibility”), context (heading, caption, nearby sentences), and constraints (max length, avoid “image of”). Request structured output (e.g., JSON) to speed review.
Measure quality: Track acceptance rate of AI suggestions and the types of edits humans make. Use that data to refine prompts.

A small-team toolchain that works

Authoring: Word/Google Docs/InDesign with accessibility checkers enabled.
Remediation: Adobe Acrobat Pro for tagging, reading order, and forms; PAC 2021 for PDF/UA validation.
Testing: NVDA (Windows, free) with Firefox; VoiceOver (macOS) with Safari; optional JAWS for parity with some enterprise users.
Contrast: Colour Contrast Analyser (desktop) to verify color choices.
EPUB: ACE by DAISY and EPUBCheck.
AI: On-device via Ollama for sensitive docs; a vetted API with a Data Processing Agreement for others.

QA with the tools your readers use

Automated checkers catch structure issues but can’t tell you if the experience is sensible. A short, repeatable manual test pays off.

Screen reader spot-check routine

Open with a screen reader on: NVDA + Firefox is a solid baseline.
Jump to the Headings list; confirm the outline is logical and levels aren’t skipped.
Tab through links; ensure labels make sense out of context.
Navigate paragraphs and lists; verify items group correctly.
Enter tables; check that header cells are announced with the right scope and that cell navigation makes sense.
Try forms; follow the tab order, ensure each field has a label, and test errors/success feedback.
Switch to Reflow and zoom to 200%+; scan a few pages for broken flow.
Where you used another language, move the cursor over that text to see if the reader switches voice.

When to choose PDF, EPUB, or HTML

PDF: Best for fixed-layout artifacts (invoices, signed reports). Requires careful tagging to be truly accessible.
EPUB: Great for long-form reading with reflow, bookmarkable chapters, and embedded semantics.
HTML: Ideal for web-first content where responsive layout and WCAG-friendly semantics are native.

For important releases, offer a dual channel: an accessible PDF plus an HTML or EPUB version. This removes layout constraints and can improve SEO and translation quality.

Shipping details that matter

Metadata: Fill XMP fields (title, author, subject, keywords). Set the default view to show the document title and bookmarks.
File name: Use a readable, versioned name (e.g., 2026-05-Quarterly-Results-Accessible.pdf).
Viewer experience: If you host PDFs in a web viewer, ensure the viewer itself is accessible and doesn’t block keyboard focus.
Plain-language abstract: Open with a short summary in everyday language; it helps many readers and screen readers alike.
Contact route: Provide a way to request formats or report issues. Listening loops make your documents better.

Troubleshooting common pitfalls

“Auto-tagging made a mess.”

That’s normal. Use auto-tagging as a starting point. Then:

Fix heading levels and nest properly under the structure root.
Convert fake lists to real L/LI structures.
Mark repetitive headers/footers as Artifacts.
Rebuild problematic tables: delete and re-tag, ensuring TH scope is correct.

“We exported from a design tool and got untagged PDFs.”

Export from the source of truth with semantic styles (Word, InDesign) instead of “print to PDF” from a layout image. If you must fix the exported file, plan extra time for manual tagging—or provide an HTML version.

“The reading order is nonsensical in two-column layouts.”

Use the Order panel to sequence content. Where the visual layout prevents a perfect order, add Bookmarks and consider repeating critical sidebars inline after their reference paragraph.

“Complex tables keep failing checks.”

Split complex grids into simpler tables, provide summaries, and if necessary, include a downloadable CSV. Screen readers aren’t spreadsheets—opt for clarity.

“Form fields won’t announce labels.”

Ensure each field has a Tooltip/label and that the tab order follows the visual order. Group related fields and add helpful error messages (not just color borders).

“Security settings block assistive tech.”

Don’t lock your PDF so tightly that reading is prevented. In Acrobat, ensure “Enable text access for screen reader devices” is allowed. Use DRM sparingly—if at all—for public documents.

Why this effort pays off

Wider reach: People who use screen readers, voice control, or high-zoom modes can finally consume your work.
Better search and translation: Tagged, well-structured content is easier for search engines and machine translation to handle.
Lower support load: Fewer “Can you send an accessible version?” tickets, and less last-minute crisis remediation.
Stronger brand: Accessibility signals care and professionalism—qualities partners and customers notice.
Faster iteration: Once your templates and AI-assisted flows are in place, each new document ships faster with fewer errors.

Start small: pick one flagship report to remediate end-to-end, establish your template and checklists, and measure the time saved on the next one. Accessibility scales when it’s part of your normal publishing workflow, not a special project.

Summary:

Accessible documents rely on semantics, reading order, alt text, contrast, and testability—not just visuals.
Build accessibility into authoring styles and templates to avoid heavy remediation later.
Remediate PDFs by fixing the structure tree, reading order, lists, tables, figures, links, and forms.
Use AI for alt text and structure suggestions with strong privacy and quality guardrails; keep humans in control.
QA with real screen readers and contrast checks; don’t rely on automated tools alone.
Ship with good metadata, bookmarks, and a plain-language abstract; offer EPUB/HTML when possible.
Treat accessibility as a repeatable workflow with templates, checklists, and metrics—not a one-off fix.

Fix Your PDFs: Practical Accessibility Remediation That Scales With AI and Good Habits

What “accessible” means for documents