Editors’ note: Today’s post is by Balamuraly Kanthavelou, Vice President at Straive. Reviewer credit to guest blogger Simon Holt.

Most publishers no longer need convincing that accessibility compliance can’t be ignored, particularly as 5,114 ADA lawsuits were recorded in 2025 alone. The European Accessibility Act and the revised ADA Title II rules are well documented at this point, and their deadlines are no mystery to most, if not all, publishers.

What is less understood is what compliance actually demands once a publisher moves past the regulatory overview and into the production reality after sitting on decades of archived content, posts, and data. There’s a clear pattern across publishers around the world: they are not struggling with awareness but with execution at volume. It’s the operational barriers that are far more concerning.

Why the Deadline Isn’t the Hardest Part

How much of our content is actually compliant today? Where are the highest-risk gaps? What will it cost to fix, and how long will it take? These are just some of the starting questions a publisher needs to ask themselves before planning a remediative course of action. Yet most organizations cannot answer these questions with any confidence, even after years of awareness that the deadlines were coming.

This is not for a lack of intent, but is rather a structural and operational challenge. Accessibility debt tends to be invisible until an organization is forced to confront it, through a compliance audit, a platform migration, or a deadline that’s suddenly looming. Traditional accessibility audits rely heavily on manual review and content sampling. These can reveal localized issues but rarely produce the complete picture of risk across an archive that spans hundreds of thousands of documents. Sampling can tell you about the one snapshot you probe, but not the lion’s share of your archive.

Because of that, accessibility initiatives tend to start with uncertainty about scope. Leadership wants to support the work, but it’s difficult to fund a program, let alone stick to a deadline, when nobody can confidently say how big the problem actually is. That uncertainty, more than the regulatory text itself, is what keeps many programs stuck at the planning stage longer than they should be.

How Legacy Content Differs

It’s worth separating two things that get conflated in a lot of accessibility conversations: making new content accessible at the point of production, and amending decades’ worth of legacy content that was never built with accessibility as a concern.

The first problem is increasingly tractable, although recent research from Michigan State University finds that nearly 75% of surveyed journals don’t explicitly request accessibility when authors submit manuscripts. Still, natively accessible workflows — where alt-text, structure, and metadata are built in during production rather than retrofitted — are more straightforward to build with embedded guidelines at the submission level.

The second problem is far more complex, and the steeper barrier for publishers to overcome at scale right now. Legacy archives accumulate inconsistencies for reasons that have nothing to do with accessibility. Different vendors over different decades, varying file formats (PDF, ePub, XML, images, application files like InDesign, 3B2, Word, datasets based on XML-based archival workflows), source files that are incomplete or missing entirely, and formatting conventions that shifted alongside every editorial change.

However, accessibility requirements have made these existing challenges more urgent for publishers to address.

Moreover, most frontlist workflows have a provision for authors to review and validate alt-text for their own work during the proofing stage. There are no such safety nets for backlist remediation because the original author may not be available to review alt-text during remediation. Where authors are unavailable, publishers will request subject matter experts to review, particularly in instances where scientific and technical images are involved.

That is added pressure on teams that have to ensure the accuracy of content not created by them, and an important dimension of the human review process in publishing.

This is also where content complexity hits hardest. A backlog of scanned PDFs with simple body text is one kind of problem, but a backlog of scholarly journal articles filled with complex and intricate figures, multi-panel charts, mathematical notations, and dense data tables is a much harder one. Academic publishing skews heavily toward the latter. The volume alone makes manual remediation far too expensive for most organizations. The added complexity slows it down, no matter how many people are involved.

The Role of AI

So, how can technology be used to address these operational challenges? First, rather than leaving publishers to guess at budget, scope, and compliance, AI can remove that friction. A model can ingest a publisher’s existing archive and run it through automated auditing to determine the scope of non-compliant content. Through that process, a prioritized inventory with a realistic timeline and cost estimate can be uncovered. This means a publisher can commit to a remediation plan with realistic numbers attached and treat accessibility as more than a one-time fix.

AI-assisted alt-text generation is also doing real work at the point of figure description, structural tagging, and metadata normalization. For straightforward content — such as simple images and consistent table structures — automated description quality has gotten good enough to be a meaningful production tool instead of a novelty. Metadata standardization is arguably where automation is furthest evolved; normalizing inconsistent legacy tagging into a coherent, WCAG-aligned structure is a problem that scales well with automation, because it’s pattern-based rather than judgment-based. These applications can also be deployed on the multimedia content within the backlists.

There are videos, recorded presentations and audio clips in the archives that are not accessible. AI can be operationalized to address this backlog by generating captions and transcripts at scale and creating audio descriptions of essential visual information. However, poor audio or video quality, unclear speech, and specialist terminology can make older recordings difficult to interpret. Human reviewers, including subject-matter experts, should validate these AI-generated outputs for accuracy and completeness.

Alt-text for complex scientific content is an entirely different story. Screen reader users encountering an ambiguous or incomplete description of a multi-panel figure or a dense equation aren’t served well by an automated description that technically exists but doesn’t actually convey the content.

Here’s where to push back on any framing of AI-driven remediation as a fully automated solution. More credible approaches pair automated generation with a confidence or quality threshold or baseline. So, content above a certain standard moves through automatically, and content below that is routed to subject-matter review before it’s considered finalized. That routing step, knowing what to flag, not just what to generate, is doing as much work as the generation itself.

The output side is also important, although it’s easy to overlook in remediation planning. Compliant content typically needs to exist in multiple formats (structured XML, ePub, and other assistive-technology-compatible formats) which means remediation isn’t just about fixing the alt-text but rather fixing the alt-text and making sure it survives correctly across every format in which the content exists.

Moreover, content validation is just as important as content generation. Automated accessibility checks, expert review, and screen reader testing should work together to verify that remediated content is accurate, navigable, and usable. Publishers can also draw on accessibility guidance and resources from industry associations such as the International Association of Scientific, Technical & Medical Publishers (STM), the Association of Learned and Professional Society Publishers (ALPSP), and the Book Industry Study Group (BISG), alongside initiatives such as the Accessible Books Consortium (ABC).

Ultimately, AI has moved to where the bottleneck sits, not eliminated it. Now, AI makes the high-volume, pattern-based parts of remediation — metadata, structure, straightforward alt-text — fast. It has not removed the need for domain expertise on technical and scientific content, and any program that treats it as a full substitute for human review on complex assets is going to produce technically tagged content that still fails its users.

Building Sustainable Workflows That go Beyond Passing Audits

Editors and publishing leaders need to pay attention to a critical distinction: whether an organization builds accessibility into how content is produced and maintained moving forward, or whether it’s running a one-off project to clear the archive.

Both produce compliant-looking output in the short term, but they are not the same thing, and the difference shows up in a few concrete places. Sustainable programs treat accessibility as an ongoing operational capability with visibility into remaining risk, cost, and progress over time, not unlike how a publisher would track any other production metric. Audit-driven programs tend to treat accessibility as a project with an end date, which means the infrastructure built to clear the backlog often doesn’t continue once the immediate pressure is alleviated. In that case, the same inconsistency problems start accumulating again with new content.

Based on experience, the practical giveaway is whether an organization can ascertain where it stands at any given moment, or only after undergoing a recent audit. Programs with clear visibility into their operations tend to catch problems early, when they are easier to fix. Those without it often rediscover the scale of unresolved issues every few years, usually when time is short and pressure is high.

Defining the Difference

It’s worth naming a few patterns directly since they define whether a program will or won’t hold up:

• Prioritize by impact, not by chronology. Working through a backlog in date order treats every asset as equally urgent, when in practice complex figures, tables, and equations carry most of the accessibility risk and should be addressed first and foremost.
• Build in human review thresholds for technical content rather than trusting fully automated output across the board, particularly for scientific and mathematical material where an incorrect or incomplete description is arguably worse than a missing one. Failing to do this creates false confidence in the audit trail.
• Treat metadata and structural consistency as foundational and not a secondary or optional afterthought. Discoverability and navigation problems caused by inconsistent metadata affect every user, not just those relying on assistive technology, and they’re often the cheapest problems to fix at scale.
• Measure progress as an ongoing capability rather than a countdown. The publishers who are going to be in good shape are not necessarily the ones with the biggest remediation budgets right now. They’re the ones who know, with reasonable confidence, where they stand today and what they still need to do.

The regulatory deadlines have done their job and got accessibility onto the executive agenda. What happens next, whether that attention turns into durable infrastructure or another compliance project that gets put on the back burner once the immediate pressure eases off, is the part that publishers are still struggling to figure out. That’s the conversation worth having now, while there’s still time to get it right.