Born-Accessible ePUB3: A Publisher's Playbook

An e-book that is accessible from the day it is published costs a fraction of one retrofitted after a complaint or a contract clause forces the issue. That is the case for born-accessible publishing: building EPUB 3 files so that structure, navigation, descriptions and metadata are right at creation, inside the normal production workflow, rather than bolted on by a remediation pass. With the European Accessibility Act now applying to e-books sold into the EU, born-accessible has shifted from best practice to baseline expectation.

EPUB 3 makes this achievable because it is built on the open web stack: XHTML, CSS, SVG and MathML. Everything the web platform can do for accessibility, an EPUB can inherit, provided the production team knows what to ask for. This post covers the specification that defines accessible EPUB, the structural and content decisions that matter most, and how certification works when you need to prove conformance to trading partners.

EPUB Accessibility 1.1, a W3C recommendation, defines what an accessible EPUB is. Rather than inventing new content rules, it maps conformance to WCAG, so a publication conforms at, for example, WCAG 2.1 Level AA, and then adds publishing-specific requirements, most importantly discoverability metadata embedded in the package file. The specification also distinguishes between the conformance of the content and the accessibility of the reading system displaying it, which keeps responsibilities clear across the supply chain.

The metadata requirements draw on the schema.org accessibility vocabulary: properties describing access modes, accessibility features such as alternative text or MathML, potential hazards such as flashing content, and a human-readable accessibility summary. Retailers and library platforms increasingly surface this metadata to buyers, so it is not bureaucratic decoration; it is how an accessible book gets found and how a publisher's investment becomes visible at the point of sale.

Reflowable EPUB, where text adapts to screen size and user settings, is inherently friendlier to accessibility: readers can resize type, change fonts and colours, and listen with text-to-speech, all without breaking the layout. Fixed layout preserves the designed page, which suits heavily illustrated children's books, art titles and some K12 materials, but it locks out many of the adaptations that make reading possible for people with low vision or dyslexia.

Our advice is to default to reflowable and to treat fixed layout as a deliberate exception that carries extra obligations: rigorous reading order in the markup, complete image descriptions, and honest metadata declaring the constraints. Some publishers produce a fixed-layout edition for visual fidelity alongside a reflowable edition for accessibility, which is legitimate when the reflowable edition is genuinely equivalent in content rather than an afterthought.

Most of an EPUB's accessibility comes from disciplined markup, the same fundamentals that govern any HTML document: real heading hierarchies rather than styled paragraphs, lists marked up as lists, tables with proper headers, and semantic inflection using epub:type and ARIA roles for structures like footnotes, glossaries and page breaks. A complete navigation document with a logical table of contents, plus a page-list when a print edition exists, lets readers move through the book rather than scroll through it.

Image descriptions deserve the same editorial investment as the captions. Alt text should be written by someone who understands the pedagogical or narrative purpose of each figure, with extended descriptions for complex diagrams. In our production work, description writing is the single largest accessibility cost line for illustrated titles, and the single largest quality differentiator in the finished book when assistive technology users review it.

Mathematics is where born-accessible publishing pays off most dramatically. Equations rendered as images are dead ends for screen reader users, while equations encoded as MathML can be voiced, navigated branch by branch, and converted to braille. EPUB 3 supports MathML natively because its content documents are XHTML, and modern reading systems and assistive technologies handle it far better than they did a few years ago.

The practical question is workflow: equations should be captured as MathML from the source, whether that source is LaTeX, Word or an XML-first pipeline, rather than rebuilt by hand at conversion time. We typically also include a rendered fallback image for older reading systems, with the MathML carrying the real semantics. Getting this right at title setup avoids the misery of re-keying thousands of equations later.

EPUB 3 media overlays synchronize recorded audio with the text using SMIL, so a reading system can highlight each phrase as it is spoken. The technique descends from DAISY talking books and serves emergent readers, language learners and people with print disabilities equally well, which is why it is common in children's publishing and literacy programmes, and increasingly requested in educational tenders.

Producing overlays well requires planning: a narration script aligned to the final text, consistent fragment identifiers in the markup, and QA that listens to the synchronization rather than spot-checking files. Overlays add cost, so they are usually reserved for titles where read-aloud is a genuine feature, but when an audio edition already exists, aligning it to the EPUB can be a surprisingly economical enhancement.

Conformance claims carry more weight when verified. Benetech's Global Certified Accessible programme audits a publisher's production output against the EPUB Accessibility specification and allows certified publishers to mark titles as such, which procurement teams in education increasingly recognize. Independent audits against EPUB Accessibility 1.1 serve a similar purpose where formal certification is not required or not yet practical for a smaller list.

The everyday tools matter as much as the certificate. Ace by DAISY automates EPUB accessibility checking and should run on every title before delivery, with the SMART tool supporting the manual evaluation that automation cannot cover. The combination of automated checks, human review with a screen reader, and accurate metadata is what makes a conformance claim defensible when a retailer or regulator asks for evidence.

Publishers rarely get to pause their list while they re-engineer production, so start where leverage is highest: fix the template and conversion specifications that every new title flows through, train description writers, and add Ace checks to the delivery gate. Then pick one frontlist title and take it through the full born-accessible process, including metadata and a screen reader review, to surface the gaps in your real workflow rather than your assumed one.

The takeaway is simple: born-accessible EPUB 3 is mostly the disciplined application of web standards you already use, governed by a specification that tells you exactly what to declare. Treat accessibility as a production specification rather than a post-production rescue, and the cost folds into business as usual while the rescue version never stops being expensive. Start with the next title you commission, not with the backlog.