What Makes an EPUB Truly Accessible Six Features That Actually Matter

Ask ten publishers what an "accessible EPUB" looks like and you'll get ten different answers. Some point to alt text and stop there. Some think reflowable text alone does the job. Some confuse a file that opens cleanly in Apple Books with a file that actually works for a reader using a screen reader or a refreshable braille display. None of these answers are wrong exactly. They're just incomplete, and incomplete is where compliance claims fall apart under scrutiny.

Frankfurt this October will have no shortage of accessibility panels, and EPUB will come up in nearly all of them. Digital rights buyers are asking sharper questions than they did even two years ago. "Is it accessible?" has turned into "which features, tested against which standard, certified by whom?" A publisher walking into that conversation needs a real answer, not a shrug and a mention of alt text.

Here's what actually separates a genuinely accessible EPUB from one that just looks the part.

Real semantic structure, not visual mimicry

A heading that's simply bold, oversized text isn't a heading to a screen reader. It's a paragraph that happens to look bigger. True accessibility starts with proper HTML5 and EPUB semantic markup: real heading tags in the right hierarchy, epub:type attributes identifying chapters, footnotes, and sidebars, figure and figcaption pairing images with their captions correctly. Assistive technology navigates by structure, not by appearance. Fake the structure and you've built a file that looks fine to a sighted reader and tells a screen reader nothing useful about how the content is organized.

Navigation that covers three things, not one

Most publishers know an EPUB needs a table of contents. Fewer know that full navigational conformance requires three separate structures working together: the TOC for chapter-level movement, a page list mapping digital locations to print page numbers for anyone working across formats, and landmarks that let a reader jump straight to the body matter, notes, or index without paging through the whole file. Skip the page list and a student using a screen reader can't cite a page number in class discussion. Skip landmarks and the file technically fails conformance even with a perfect TOC sitting right there.

Alt text that describes, not decorates

Every meaningful image needs an alternative. That much is well known. What trips publishers up is depth. A stock label like "chart" or "diagram" checks a technical box and helps nobody. A reader needs to know what the chart shows: the trend, the comparison, the number that matters. Complex images, the kind found throughout STM and technical publishing, often need a longer description beyond what fits in a standard alt attribute, linked separately so the detail is there without cluttering the reading experience for everyone else.

Reading order that matches what a sighted reader actually sees

A page can look completely normal and still read in the wrong order underneath. Sidebars before body text. Captions inserted mid-sentence. Footnotes appearing before the paragraph that references them. None of this shows up by glancing at a rendered page. It shows up when you test with an actual screen reader, or run the file through a validator like Ace by DAISY. Reading order problems are some of the most common issues we find in EPUB conversions, especially in files migrated from print layout software that never had to think about linear structure.

MathML wherever equations live

An equation rendered as a flat image is invisible to assistive technology and useless to anyone who needs to resize or re-flow the content. MathML preserves the actual mathematical structure so it can be read aloud correctly, not just displayed. This matters most in STM and higher education content, where a single physics or statistics title can carry hundreds of equations, and where getting this wrong means the entire subject area of the book is effectively unreadable for some students.

Metadata that tells the truth about what's inside

This is the feature publishers skip most often, and it's the one buyers check first. EPUB Accessibility 1.1 requires specific metadata fields declared in the package file: accessMode describing how the content is consumed, accessibilityFeature listing what's actually present, accessibilityHazard flagging anything like flashing content, and a conformsTo statement naming the WCAG level the file meets. Without this, a distribution platform has no reliable way to classify the file, and a procurement team evaluating it for institutional purchase has nothing to check it against except taking the publisher's word for it. Word alone doesn't survive a compliance audit.

One more thing: fixed layout has real limits

Fixed-layout EPUBs, common in illustrated and children's books, can be built with a reasonable degree of accessibility, but they can't fully satisfy every requirement the way reflowable text can. WCAG's Reflow criterion depends on content that can adapt to a reader's chosen text size and display settings, something fixed layout inherently restricts. That doesn't mean fixed-layout titles should skip accessibility work. It means publishers should know going in what the format can and can't deliver, and document that honestly rather than promising full conformance a fixed canvas can't provide.

Why this matters walking into Frankfurt

None of these six features work in isolation. A file with flawless alt text and broken reading order still fails a reader using a screen reader. Perfect semantic markup with no accessibility metadata still gets skipped by a procurement system built to check for declared conformance. Genuine accessibility is the sum of all of it, verified against WCAG 2.1 AA and EPUB Accessibility 1.1, tested with the same tools a certifier would use, not just eyeballed once and called done.

That's the standard behind every file that passes through S4Carlisle's accessibility pipeline. The NINJA AI Ecosystem runs first-pass detection across semantic structure, navigation, alt text, reading order, and metadata, and trained reviewers verify the result before anything ships, backed by Benetech GCA certification. A publisher bringing an EPUB catalogue to Frankfurt this year doesn't need to guess whether it holds up. It needs a partner who has already tested it against every one of these six features, not just the easy ones.

S4Carlisle is Benetech GCA certified and runs EPUB accessibility remediation at scale through the NINJA AI Ecosystem. Visit us at Frankfurt to talk through what your files actually need.