Re-export or repair? The route to a verified PDF – for existing documents too, with an honest distinction between what automation does and what needs manual work.
An accessible PDF is not created with a single click, but through a series of steps that can be planned well. Whether you are creating a new document or improving an existing one: it all starts with a decision.
Decide first: re-export or repair?
- Source document available (Word, InDesign, Writer) and still up to date? Then correct it in the source document and re-export it with tags. This is almost always the cleanest and most lasting route – instructions in the 10-point guide for Word.
- Only the PDF available? Then the PDF itself is repaired: add or correct the structure, fill in missing information, check.
- Scanned document (the pages are only images)? First create a text layer using text recognition (OCR), then continue as in point 2.
- Legacy documents? Public bodies may exempt documents that were published before 23 September 2018 and are not needed for any ongoing administrative procedure – an exemption provided for by the Web Accessibility Directive (EU) 2016/2102. They then belong in the accessibility statement (see BITV 2.0 for municipalities – BITV 2.0 being Germany’s federal accessibility ordinance for IT).
The steps in detail
- Real text instead of images. If the text cannot be selected, the document is a scan. Without OCR it remains silent for screen readers.
- Tags and reading order. All content needs a tag, and the order in the tag tree must match the reading order – the most common source of errors on multi-column pages, boxes and margin notes. Headers and footers, page numbers and decorative elements are marked as artifacts.
- Headings. Mark up headings as H1, H2, H3 … without skipping levels. They are the table of contents for screen reader users.
- Lists. Tag bulleted and numbered lists as lists (L, LI, Lbl, LBody), not as paragraphs with dashes.
- Tables with header cells. Data tables as Table with rows and cells; header cells as TH with a scope (column or row). Do not misuse tables as a layout tool.
- Alternative text. Every informative image gets a short description of what it conveys; decorative images become artifacts.
- Language. Set the document language (e.g. en-GB) and mark up passages in other languages separately.
- Title. Set a meaningful document title and configure the document to display the title instead of the file name.
- Bookmarks. For longer documents, generate bookmarks from the headings.
- Forms. Fillable fields with a label (tooltip), a sensible tab order and a link in the structure; make mandatory fields recognisable.
- Contrast and colour. Text at least 4.5:1 (large text 3:1); do not convey information through colour alone. Contrast can rarely be changed in the PDF afterwards – usually only the source document helps here.
- Checking. By machine with a PDF/UA validator such as veraPDF or PAC, then the human checks: listening with a screen reader, tab test, looking at alternative text and reading order (how to test it yourself).
- PDF/UA identifier. Only set it once the check passes – otherwise it is a false claim of conformance.
- Accessibility statement. Honestly name what is not (yet) accessible – with the reason and a timetable (guide).
Where automation helps – and where people are needed
- Easy to automate: OCR, title and language, title display, bookmarks from headings, tab order, embedding of standard fonts, link descriptions, formal repairs to the tag tree, the PDF/UA identifier after a passed check.
- Automation with a proposal, a human decides: tags for previously untagged documents, alternative text, table structure, form fields – here software provides a proposal that has to be reviewed.
- Only people: whether alternative text is correct in content, whether the reading order is logical, whether headings and link texts are understandable, whether the contrast in the design is sufficient.
Software that makes any PDF fully automatically and guaranteed compliant does not exist. What is realistic: the machine does the routine work, a human answers the questions of judgement – with considerably less effort, because only those are left.
How it works in DokAudit
- Single check: upload a PDF or Word file. Word is converted into a tagged PDF; PDFs are checked, every finding is explained in plain language, repairs are selected with a tick. If the structure is missing or only rudimentary, DokAudit generates it automatically (headings, paragraphs, lists, tables, links), multi-column layouts with AI structure recognition; scans first receive an OCR text layer. Only headings set in bold are marked up as H1–H3.
- Conformance loop: when repairing, DokAudit checks with veraPDF, fixes the open rules in a targeted way and checks again – several rounds, until the PDF/UA-1 check passes or only points remain that a human has to clarify. A round that improves nothing is discarded; the report states what remains and why.
- Hub with full automation: in the document library, uploaded documents are checked, repaired and re-checked without an approval click – practical for larger collections.
- Form tool: the “Make form fillable” tool detects fields in flat application forms as a proposal; after your correction, fillable fields are created with tooltip, tab order and – for tagged documents – a link in the structure.
- UA-2 version: if the repaired version passes the PDF/UA-1 check, a PDF/UA-2 version (PDF 2.0) can additionally be produced – only provided if it passes the UA-2 check (more on PDF/UA-1 and -2).
What remains afterwards is shown explicitly in the report: approving AI proposals for alternative text, listening to the reading order on a sample basis with a screen reader, and, in case of doubt, resolving complex tables and contrast issues in the source document.
Sources:
As of: 09/2026. This article provides a general overview and is not legal advice. The statutory and standards texts in force are authoritative; in individual cases the law of a German state may differ.