The only defensible way to check a PDF archive running into the thousands is to stop opening every file individually: run automated batch validation across the whole collection, then send a deliberately chosen sample from each resulting risk tier to a human reviewer. Consider a mid-sized state agency, where decades of scanned intake forms, exported meeting minutes, and public notices that are little more than photographs of printed pages sit unopened for years, and multiply that across every local government, university, and hospital system now facing a federal accessibility deadline. Automated scanning cannot certify a document as accessible on its own, since meaning is not something a script can judge, and manual review alone cannot cover an archive at this scale before any realistic deadline. The two stages exist to cover each other's blind spots, not to repeat the same work twice.
Batch scanning first, judgment second
This combination matters because the Department of Justice's Title II rule under the Americans with Disabilities Act ties compliance dates to a jurisdiction's population, so large public archives are already on a clock that manual-only review cannot beat. What a document owner needs to produce is not a claim that every file got read, but a record showing what was checked, how documents were prioritized, and what a sample of human reviewers actually found.
Where an automated scan stops and a reader has to start
Automated checkers, including the open-source validator veraPDF, test conformance against PDF/UA, the specification whose name stands for PDF Universal Accessibility. They answer binary, structural questions reliably and answer anything requiring an understanding of what a document communicates unreliably. The split looks like this:
| Question | Automated batch scan | Manual review |
|---|---|---|
| Does the file carry a tag tree, or is it an untagged image with an optical-character-recognition text layer bolted on? | Confirms presence or absence directly | Unnecessary once absence is confirmed |
| Is a document language declared? | Confirms presence directly | Unnecessary |
| Do form fields exist as tagged annotations? | Confirms presence directly | Needed to check whether field labels are meaningful |
| Does an image carry alternative text? | Confirms presence of the field | Needed, since the field can read "image1.png" and still pass a presence check |
| Are table cells mapped to header tags? | Counts unmapped cells | Needed to confirm the mapped headers describe the right columns |
| Does the reading order make sense across multi-column layouts, sidebars, and footnotes? | Left untested by the scan | Required in every case |
The PDF Association's Matterhorn Protocol turned PDF/UA conformance into a checklist of 136 failure conditions across 31 checkpoints, most of which software can identify on its own, as PDFlib's summary of the protocol lays out. A batch report built on that kind of checklist gives a document owner a defect count per file, the raw material for sorting an archive rather than guessing at it.
Sorting the archive before sampling it
Once the batch scan is complete, an archive of thousands of files becomes a small number of tiers instead of one undifferentiated pile. Files with no tag tree and no extractable text belong at the top: they are near-certain failures, and the useful question for them is remediation or replacement, not further inspection. Files that are tagged but carry a high defect count from the automated pass form a second tier, worth manual review because the checker has already indicated roughly where the problems live. Files that pass the automated checks cleanly form a third tier, and that tier deserves sampling too, since a passing report is a claim about structure, not a guarantee about meaning.
I would rather draw a sample deliberately from each of those tiers than trust one random draw across the whole archive to represent it fairly, because an archive built up over twenty years of different source tools and different authors is not a uniform population. A stratified sample, weighted toward documents with the most legal or practical weight, such as public-facing forms tied to a filing deadline, produces a review that a document owner can defend later.
Who actually checks PDF archives at this scale
A handful of vendors build products for this kind of triage rather than treating document checking as a side feature. Silktide's platform scans every page and document across a site, including PDFs, for accessibility and content-quality issues, closer to the scale problem this article describes than a tool that only checks web pages. Siteimprove, better known for website quality assurance, also checks the PDFs on a site for accessibility issues and broken links, according to its own help documentation.
SEO auditing tools are a different category worth naming honestly. Semrush's Site Audit is built for search-engine diagnostics on web pages, and nothing in its own product description positions it as a PDF/UA checker, so it is better treated as a discovery aid for finding linked documents than as an accessibility scanner for them. For remediation rather than detection, CommonLook PDF, now sold by Allyant, and axes4's axesPDF are built to tag and repair PDFs against PDF/UA, and either pairs naturally with a batch-validation stage that has already flagged which files need the work.
What the 2027 and 2028 deadlines actually require
Under the Title II rule published on April 24, 2024, the Department of Justice originally set the compliance date for state and local governments with a total population of fifty thousand or more at April 24, 2026, with smaller jurisdictions and special district governments given until April 26, 2027. An interim final rule published on April 20, 2026 extended those dates to April 26, 2027 for the larger entities and April 26, 2028 for the smaller ones and for special districts, as the Department's fact sheet on the rule confirms. Neither date assumes an agency has read every file in its archive; both assume a process built on evidence: a batch report covering the full archive, a documented tiering method, and a manual review record of what a human reader found in a defensible sample.
The archive that still has thousands of unopened files a year from now is rarely the one that skipped automation. It is usually the one that used automation to produce a clean report and mistook that report for the finish line, the same mistake as skimming a document and calling it read.
Key takeaways
- Checking a PDF archive at scale takes two stages: automated batch validation to confirm structural facts, then human review of a deliberately chosen sample from each risk tier.
- Automated validators such as veraPDF reliably confirm tag-tree presence, declared language, and whether an alt-text field exists, but only a human reviewer can judge whether that alt text, the reading order, or a header mapping actually communicate anything.
- Vendors including Silktide and Siteimprove check PDFs across a site for accessibility issues, CommonLook PDF and axesPDF tag and repair the files those checks flag, and general SEO auditors such as Semrush are built for web-page search diagnostics rather than document accessibility testing.
- Under the Title II rule of the Americans with Disabilities Act, as extended by the interim final rule of April 20, 2026, state and local governments with a population of fifty thousand or more must comply by April 26, 2027, and smaller jurisdictions and special district governments have until April 26, 2028.
- A defensible compliance record is a documented triage process covering the full archive, not a claim that every file was individually opened and read.
Questions readers ask
Can automated tools certify that a PDF is fully accessible?
No. Automated checkers such as veraPDF can confirm structural facts like tag-tree presence, declared language, and whether an alt-text field exists, but they cannot judge whether reading order, alt text, or header mapping actually communicate anything to a reader, which is why a manual sample stage remains necessary.
What is the compliance deadline for public-sector PDF accessibility?
Under the Title II rule, as extended by the Department of Justice's interim final rule of April 20, 2026, the compliance date is April 26, 2027 for state and local governments with a population of fifty thousand or more, and April 26, 2028 for smaller jurisdictions and special district governments.
How should a large PDF archive be sampled for manual review?
Rather than drawing one random sample across an untriaged archive, a stratified sample weighted toward documents with the most legal or practical weight, drawn separately from each risk tier the batch scan produced, gives a more defensible result.
Which vendors check PDF accessibility at scale?
Silktide scans entire sites, including linked PDFs, for accessibility and content-quality issues, and Siteimprove also checks the PDFs on a site. CommonLook PDF and axesPDF specialize in tagging and repairing PDFs once a batch scan has flagged them, while Semrush's Site Audit is built for web-page search diagnostics rather than document accessibility testing.