A recent Carnegie Higher Ed article made a claim worth sitting with. Website bloat in higher education, the piece argued, isn’t a technology problem. It’s a governance problem. Colleges pile up thousands of PDF files on their public sites because saving a Word doc as a PDF is the easiest path for busy staff, not because PDFs are the best way to serve their students.
At AndiSites, we agree. And we would add something else. The PDF question is not just a governance issue. It is also a question of timing, budget, and legal exposure. The U.S. Department of Justice’s Title II web rule sets two compliance dates: April 26, 2027 for entities with populations of 50,000 or more, and April 26, 2028 for smaller entities and special district governments. For most colleges and public agencies, a document library that has grown unmanaged for a decade cannot be remediated line by line in the time or the budget available.
So the question is not “should we convert PDFs to HTML?” That answer is often yes. The real question is: which PDFs, in what order, converted by whom, and reviewed how? That’s where most institutional teams get stuck, and it’s where partners like AndiSites can help.
The AI “Conversion” Wave Is Already Here
AI-assisted document remediation has existed for years. Continual Engine’s PREP platform describes AI-enabled remediation dating back to 2019, with PDF and Office-to-HTML conversion features added by 2021. What changed in 2026 is not that AI conversion was invented. What changed is availability, economics, scale, and institutional adoption. Some state governments are building AI-driven tools that claim to convert PDFs to accessible HTML at scale and embed them directly into government publishing workflows–albeit with explicit warnings that converted pages may still require accessibility verification and that some documents may fail the automated quality threshold entirely.
What this means practically: the raw conversion work is becoming commodity. It’s fast, it’s cheap, and for straightforward informational content, it works. That is genuinely good news for accessibility, and it’s good news for institutional budgets that have been dreading five and six figure remediation quotes for years.
But it also means the question changes. If a tool can convert your program guide to HTML in three minutes, the leverage is no longer in “can we do it.” The leverage is in “can we do it right, in the right order, integrated with our existing website, with real quality assurance behind it.”
Why AI Conversion Still Needs Human Judgment
The AI does the raw structural conversion well. What it does not do well is finish. The state government example above is a case in point: the tool acknowledges the file still needs review, but there is no human in the loop by default. Someone has to catch the duplicate content, the header artifacts, the tables that look right in HTML but read backwards to a screen reader. That’s the layer where accessibility either happens or fails.
- AI should not be relied on to decide which PDFs should stay PDFs. A signed board resolution isn’t a candidate for HTML conversion. A print-ready program viewbook isn’t either. A fillable enrollment form serving specific accessibility needs may or may not be, depending on how it’s used. Those calls take a human with context.
- AI should not be relied on to judge whether the output is actually accessible to a screen reader user. Ninety percent accuracy on structural detection is impressive, but it’s not the same as ninety percent accuracy on lived experience. A page with correctly tagged headings can still read as gibberish if the reading order is wrong, and no automated checker will catch that reliably.
- AI does not integrate the resulting HTML into your existing website on its own. WordPress themes, navigation structures, categories, tagging, permalinks, and SEO all require a human deciding how the converted content actually lives on your site. A 300-page PDF broken into 40 HTML pages is a content architecture problem, not a conversion problem.
- AI does not advise you on the governance layer. Once the initial backlog is converted, someone has to make sure new PDFs don’t immediately start piling up again. That means adopting a PDF-by-exception policy, training content owners, and building a review calendar. Software does not do that. People do.
How We Think About It
At AndiSites, we approach the PDF question in three tiers, and clients pick the one that fits their situation:
- PDF-Only Remediation is for institutions that need to bring existing PDFs into compliance without changing format. Signed documents, print-ready collateral, and interactive fillable forms usually belong here. We tag, structure, and QA these files against WCAG 2.1 AA and PDF/UA-1 standards. This is our original service line, refined across engagements with technical and community colleges.
- PDF-to-WordPress is our newer offering, built for institutions ready to move informational content into HTML pages that live natively on their websites. We handle the strategy audit (which files convert, which stay PDFs); run the AI conversion pass through a vetted vendor; QA the AI output against real accessibility standards; and publish the resulting HTML into WordPress with your theme, navigation, and taxonomy handled properly. Clients who want to leverage AI conversion without owning the full workflow themselves land here.
- Hybrid combines both. Clients with institutional document libraries often end up here. A library with 300 PDFs might have 50 legal documents that stay as PDFs, 150 program guides that convert to HTML, and 100 fillable forms that get remediated in place. Our strategy audit assigns each document to its appropriate tier, and delivery runs through the same integrated project.
What Governance Actually Looks Like
Carnegie’s core argument is that PDF bloat starts with policy, not technology. We would take that further. Governance that prevents future bloat has to include four things:
- A PDF-by-exception default. New content publishes as HTML unless a specific category applies (formal legal, print-ready, interactive form).
- A named accessibility coordinator or lead. Not a committee. One person or role that reviews exception requests and owns the ongoing document review calendar.
- A published lifecycle policy. Every PDF carries a review date. Program guides revisit annually. Handbooks every eighteen months. Legal documents every three years or when regulation changes.
- Real enforcement. Documents that violate policy get flagged and removed until remediated. Repeat violations get addressed through the normal institutional performance-review processes, not left to accumulate.
None of this is complicated, but it’s too easy to forget or ignore. AndiSites helps clients set up governance as part of their engagement so what we deliver stays accessible over time.
What Should You Do Next?
If your organization has a DOJ Title II deadline coming; a document library that’s been growing without oversight for a decade; and an available budget that just can’t accommodate hand-remediating every PDF you have on file, you’re in the group AndiSites’ services are built for. New AI conversion tools make the math work in ways it didn’t even a year ago. But AI still needs a human strategy layer above it and a human QA layer below it. That’s where we come in.
If you’d like a free Zoom to talk through what your library looks like and which tier fits, please let us know. No heavy sell or obligation, we promise–just help understanding your options as the April 2027 and April 2028 deadlines gets closer.
AndiSites is here to help you with accessibility, PDFs, and compliance for the 2027 and 2028 deadlines.