Accessibility Cannot Stay at the End of Production

A figure caption can be beautiful and still leave a reader out. A PDF can look finished and still be almost unusable with a screen reader. A journal can publish open access articles, with no paywall and a clear license, while quietly making the content harder to use for exactly the readers open access claims to serve.

That tension is becoming harder for journal offices to treat as a specialist afterthought. On August 17, 2026, DOAJ published a practical account of how Penn State University Libraries Open Publishing uses the OA Journals Toolkit across a growing portfolio of open access journals: https://blog.doaj.org/2026/08/17/putting-the-oa-journals-toolkit-to-work-in-library-publishing/. The most telling detail was not a new platform feature or a legal warning. It was the ordinary operational use of accessibility remediation resources, including alt-text guidance for authors, to update journal instructions and support editors.

That is the point. Accessibility work is becoming journal work. Not only web work. Not only PDF repair. Not only a vendor ticket after the issue is assembled. For editors, managing editors, production leads, and library publishers, the question is shifting from whether accessible formats are desirable to where the work enters the manuscript lifecycle.

The Figure Arrives Before The Standard

Most accessibility failures begin before production sees a file. A submission arrives with six figures, three tables rendered as images, a visual abstract, a supplementary slide deck, and no usable description of what a non-sighted reader needs to know. By the time the article is accepted, copyedited, typeset, and exported, everyone knows the article is late. Asking production to invent meaningful alt text at that point is not a quality system. It is a rescue mission.

Author guidelines are therefore not a courtesy page. They are the first accessibility control. A journal that expects image descriptions should say so at submission, explain the difference between a caption and alt text, and give examples by content type. A microscopy image, a map, a Kaplan-Meier curve, a flow diagram, a chemical structure, and a conceptual illustration do not need the same descriptive strategy. The author usually knows the scientific point of the image better than a downstream remediator does.

The OA Journals Toolkit example matters because it shows a library publishing program using shared open resources to make this work teachable. That is especially useful for smaller journals and scholar-led programs that cannot maintain a dedicated accessibility department. A checklist is not enough, but a shared language gives editors something concrete to ask for before the manuscript becomes a production problem.

Compliance Language Is Getting More Concrete

The standards environment is also becoming less abstract. W3C explains that WCAG 2.2 is the latest version of the Web Content Accessibility Guidelines and that WCAG applies beyond ordinary web pages, including documents and other non-web information and communications technologies: https://www.w3.org/WAI/standards-guidelines/wcag/. W3C also notes that WCAG 2.2 was approved as ISO/IEC 40500:2025 and encourages use of the latest version.

STM has been working on alt-text accessibility for scholarly images through an active Task & Finish Group. Its page links the work to WCAG 2.1 AA expectations, the April 2026 ADA Title II update for U.S. institutions, and the European Accessibility Act, while developing a shared image-type taxonomy for short descriptions: https://stm-assoc.org/what-we-do/strategic-areas/standards-technology/alt-text-accessibility-standards/. The taxonomy work is voluntary, but the operational signal is not. Scholarly images are too varied for generic alt-text prompts to carry the load alone.

Large publishers are moving from statements to workflows. Taylor & Francis says that beginning March 2026, all content published in its journals includes image descriptions, with AI-generated descriptions for meaningful images and human review by typesetting vendors: https://taylorandfrancis.com/about/corporate-responsibility/accessibility-at-taylor-francis/. Smaller publishers do not need to copy that exact model. They do need to notice what it implies: accessibility is being routed through intake, production, vendor governance, and platform display, not just through a policy paragraph.

PDF, HTML, And EPUB Are Different Promises

A common mistake is to treat the article as one object with several wrappers. Accessibility exposes the weakness in that thinking. HTML may have clear headings and link text. The PDF may have broken reading order. The EPUB may reflow well but lose a table relationship. A supplementary file may be a scanned image with no text layer. The article is not accessible because one format passes a quick check.

PDF work is especially easy to underestimate because visual polish can hide structural failure. The PDF Association released a 2025 best practice guide on conforming to both PDF/A for archiving and PDF/UA for accessibility, explicitly framing the challenge as long-term preservation and usability together: https://pdfa.org/conforming-to-both-pdf-a-and-pdf-ua-a-new-best-practice-guide/. For journals, that pairing is important. An archive that survives but cannot be navigated is not a fully successful archive.

This is why accessibility should be tested by format and by article component. Headings, tables, formulae, references, figures, supplements, video abstracts, and correction notices all have different failure modes. If the journal publishes in HTML and PDF, staff should know which accessibility checks belong to each output and which defects require a production correction before publication.

AI Can Help, But It Cannot Own Context

AI will be part of this work. It is already being used to draft image descriptions, caption video, and inspect documents. That can reduce the blank-page problem for authors and production teams. It can also create plausible but scientifically thin descriptions if the model does not understand why the image matters in the article.

The right question is not whether AI may write a first draft of alt text. The question is who verifies the description against the argument of the paper. A heatmap may show color clusters, but the meaningful description depends on the variables, comparison groups, uncertainty, and claimed interpretation. A pathology image may need scale, staining, and diagnostic feature context. A flow diagram may need the decision points, not the shape of the boxes.

Journal policy should therefore separate assistance from accountability. Authors can be asked to provide or review descriptions. Production can run structural checks. Vendors can generate drafts under clear data and confidentiality terms. Editors should know when missing or misleading accessibility material affects the published record. The goal is not perfect prose in every alt attribute. The goal is a documented path from visual evidence to accessible representation.

Backfiles Need Triage, Not Shame

No journal with a long archive can repair everything at once. Older PDFs may lack tags, scanned pages may have poor OCR, figures may have no source files, and discontinued journals may sit on platforms nobody planned for accessibility. Treating the backlog as a moral failure is not useful. Treating it as an inventory problem is.

Start with high-use content, teaching-heavy articles, clinical or policy material, special issues that still attract readers, and pages where users have requested accessible versions. Then separate remediable defects from structural limits. Some fixes can be handled with OCR, headings, table markup, link text, and alt descriptions. Others require a replacement file, an alternate format, or a clear request route when full remediation is not immediately feasible.

The backfile plan should also protect new content from becoming tomorrow's backlog. If a journal keeps publishing inaccessible figures while slowly repairing old PDFs, it is filling the hole while digging beside it. New submissions need the stronger workflow first; archive remediation can then proceed as a managed program instead of a permanent emergency.

Practical Takeaway For Journal Leaders

Pick one article accepted in the last month and run an accessibility walk-through from submission to publication. Check whether author guidelines requested image descriptions, whether the submission form captured them, whether copyediting reviewed them, whether production preserved them in HTML and PDF, whether tables and headings remained navigable, whether supplementary files were considered, and whether staff could answer a reader request for an accessible version.

Then fix the first broken handoff. Do not begin with a grand accessibility strategy if the journal cannot yet collect alt text at submission or preserve headings in PDF export. The immediate leadership task is to move accessibility upstream, assign review points, and make each output format accountable. Open access removes the price barrier. Accessible production removes another barrier that readers encounter after the paywall is already gone.