Conference 19–20 September 2023 📍 Boston, Massachusetts, USA

PDF Days North America 2023 — Accessibility, AI, and the Next PDF Standard

PDF Days North America 2023 was held in Boston, Massachusetts on 19–20 September 2023. Sessions covered the state of PDF accessibility compliance under tightening US and EU deadlines, early discussions about AI-assisted PDF processing workflows, and the development roadmap for PDF 2.1 and PDF/UA-2.

Official event page →

Key takeaways

Three years of Section 508 enforcement data confirmed: document age predicts inaccessibility; complex forms require disproportionate remediation effort; procurement requirements are the most effective compliance lever.
PDF/A-4 was not yet covered by NARA guidance as of October 2023, limiting its use for federal permanent records — agencies were advised to evaluate tools but defer archive migration.
Generative AI (multimodal models) was demonstrably useful for alt text generation and reading order correction, with reported 40–60% remediation cost reductions in early deployments.
The consensus position: AI-assisted accessibility tools are production accelerators that require human review, not automated compliance guarantees.

PDF Days North America 2023 was held in Washington D.C. in October 2023, organised by the PDF Association. The 2023 North American conference came three years after the August 2020 deadline by which US federal agencies were required to make all new ICT (including PDFs) Section 508 compliant, and two years after the initial 2021 retrospective. The additional two years of enforcement experience had produced substantially more data about what sustained Section 508 PDF compliance actually looks like in practice. The conference also marked the introduction of PDF/A-4 to the North American audience in the context of federal archiving requirements, and addressed generative AI's emerging role in PDF accessibility workflows — a topic that had moved from speculative to demonstrably practical in the 18 months since PDF Days Europe 2023.

Three years of Section 508 enforcement: data-driven findings

By 2023, US federal agencies that had gone through multiple Section 508 audit cycles could provide empirically grounded findings rather than first-impressions. Sessions reviewed findings from agency remediation programmes: Document age is the strongest predictor of inaccessibility — Documents created before 2019 (before the Section 508 Refresh compliance deadline) showed far higher rates of accessibility failures than post-2019 documents. Legacy document libraries were the dominant remediation burden. Agencies that had focused remediation effort on recently accessed documents — prioritising by download frequency — found better return on investment than approaches that tried to remediate entire libraries chronologically. Form complexity drives remediation cost — Complex multi-page forms (tax forms, benefits applications, procurement forms) required disproportionate remediation effort. A 50-page complex form with tables, branching logic, and signature fields could require 10× more remediation effort than a 50-page prose document. Sessions provided guidance on when to retire complex PDF forms in favour of accessible HTML forms. Procurement progress — The approach of specifying PDF accessibility requirements in vendor contracts (discussed at PDF Days North America 2021) had produced measurable results. Agencies that had included specific PDF/UA conformance requirements in software procurement contracts were receiving more accessible PDF output from vendors than those relying on general accessibility statements. OCR quality variability — Federal agencies managing scanned legacy documents were finding significant quality variation in OCR text layers. Documents scanned from microfilm, faded originals, or dense typewritten forms had higher OCR error rates that affected both text searchability and screen reader output. Sessions covered OCR accuracy measurement and the decision criteria for when to perform manual OCR correction versus accepting automated output.

PDF/A-4 and federal records management

PDF Days North America 2023 introduced PDF/A-4 to US federal document professionals who were primarily familiar with PDF/A-1 and PDF/A-2 from NARA (National Archives and Records Administration) guidance. NARA guidance lag — NARA's format guidance (Bulletin 2014-04 and subsequent updates) had endorsed PDF/A-1 and PDF/A-2 for permanent records. As of October 2023, NARA had not yet issued guidance explicitly endorsing PDF/A-4. Sessions discussed the practical implication: agencies could not designate PDF/A-4 as a NARA-approved format for permanent records without NARA concurrence, even though PDF/A-4 was a published ISO standard with superior technical capabilities. PDF/A-4f and federal e-invoicing — The US federal government's transition to electronic invoicing (under OMB guidance on invoice automation) had created interest in PDF/A-4f as a format for hybrid human/machine-readable invoices — the federal equivalent of the EU's Factur-X requirement. Sessions reviewed the technical requirements for this use case and compared with PDF/A-3 approaches already in use. Migration from PDF/A-1 archives — For agencies with large existing PDF/A-1 archives, the migration question was becoming relevant. PDF/A-1 documents would continue to be valid and accessible indefinitely — migration was not urgent. But new document creation could begin targeting PDF/A-4 once tool support and NARA guidance were in place. Sessions provided a practical assessment: begin tool evaluation and procurement planning, but do not execute archive migrations until NARA guidance is clear.

Generative AI in federal PDF accessibility workflows

By late 2023, generative AI tools were actively being evaluated for PDF accessibility use cases in federal agencies. Sessions provided the first structured review of deployment experiences: AI-generated alt text — Multimodal AI models (GPT-4V, Claude, Gemini) could generate image alternative text for figures in PDFs with reasonable quality for photographs and charts. For technical diagrams and engineering drawings, accuracy was lower and human review remained necessary. Cost per figure was dramatically lower than manual alt text authoring. Several agencies reported using AI-generated alt text as a first draft reviewed and corrected by accessibility specialists — reducing per-document cost by 40–60% compared to fully manual approaches. Automated reading order correction — AI layout analysis models trained on annotated PDF datasets could identify and correct reading order errors in multi-column documents with higher accuracy than rule-based approaches. Early production deployments in document management systems were reducing per-page remediation time significantly. AI-generated document summaries for skip navigation — Some approaches used LLMs to generate brief document summaries placed at the beginning of long PDFs, providing a functional equivalent of skip-navigation for screen reader users navigating to specific sections. While not a PDF/UA requirement, this was valued by agencies as a user experience improvement beyond minimum compliance. Limitations emphasised — Sessions were careful to note that AI-assisted accessibility was a production accelerator, not a compliance replacement. Automated tools (AI or otherwise) cannot verify all PDF/UA requirements — particularly those requiring human judgment about semantic correctness, appropriate use of language, and the coherence of reading order in complex content.

Frequently asked questions

What are the Section 508 requirements for PDFs in 2023 and beyond?

The 2017 Section 508 Refresh (the "Revised 508 Standards" from the US Access Board) requires ICT to meet WCAG 2.0 Level AA as the baseline. For PDFs, the applicable PDF-specific WCAG techniques are in the W3C's PDF Techniques for WCAG 2.0 series (PDF1 through PDF23). Key requirements include: text alternatives for images (PDF1), bookmarks for long documents (PDF2), correct tab order in forms (PDF3), document language declaration (PDF16), heading structure (PDF9), table header associations (PDF6), and form field labels (PDF10). Note that while the 508 Refresh references WCAG 2.0, many agencies are now targeting WCAG 2.1 or WCAG 2.2 for new content as the agency interpretation of "substantial equivalence". Check the Access Board's ICT Testing Baseline for PDFs for the current federal government testing methodology.

What is NARA and why does its guidance matter for federal PDF formats?

The National Archives and Records Administration (NARA) is the US federal government's official records management authority. It sets policy on what file formats are acceptable for permanent federal records — records transferred to NARA at the end of their agency retention period. NARA's Bulletin 2014-04 (Format Guidance for the Transfer of Permanent Electronic Records) lists acceptable formats; PDF/A (specifically PDF/A-1 and PDF/A-2) is on the preferred formats list. Agencies cannot unilaterally decide to use non-listed formats for permanent records — they need NARA concurrence. Because NARA guidance typically lags commercial format adoption by several years (PDF/A-1 was published in 2005; NARA formally endorsed it in 2014), new standards like PDF/A-4 require agencies to either wait for NARA guidance or seek agency-specific NARA concurrence before adoption.

Can AI tools create fully Section 508 compliant PDFs automatically?

Not currently and likely not in the foreseeable future, for several reasons: (1) Some PDF/UA and WCAG conformance requirements demand human judgment — for example, determining whether alternative text accurately describes an image's purpose, or whether reading order matches the author's intent, or whether a complex form's branching logic is accessible. (2) AI tools make errors, particularly on unusual document layouts, domain-specific diagrams, and fine-grained semantic structure. Error detection and correction requires human review. (3) PDF generation from source documents is often the right intervention point — improving the authoring tool or template produces accessible PDFs from the start, without post-hoc remediation. AI tools are most valuable as components in a human-supervised workflow: auto-generate tags and alt text, have a specialist review and correct, then validate with PAC/veraPDF before publication.

Try it in your browser

The PDF tools and standards discussed at this event are available free — no account needed.

Fill PDF form Edit PDF Annotate PDF Open editor