Tagged PDFs: the invisible layer that decides who can read your documents
Why two identical-looking PDFs can be night and day for a blind reader: what tags are, how reading order really works, what artifacts do, and how remediation adds the layer without changing a pixel.
Editorial
Open two copies of the same report — identical on screen, page for page. A screen-reader user opens both: the first reads cleanly, heading to heading; the second is word salad interrupted by page numbers. The difference is a layer you can't see.
A PDF is two documents
PDF began as a printing format: a script recording where every stroke of ink goes. Nothing in that script says this line is a heading or read the left column first. That meaning lives — when it exists — in a second, invisible document inside the file: the tag structure, a tree of labels (Heading 1, Paragraph, Table, Figure) arranged in reading order, standardized by PDF/UA.
Screen readers navigate the tags, not the ink. Which yields the strange, useful truth of this field: accessibility repairs never need to change how a PDF looks — they build the second document to match the first.
The parts of the invisible layer
- Tags — the skeleton. Headings enable jump-navigation; table tags enable reading by row and column. "Document is not tagged" tops every audit because most software never writes them.
- Reading order — the tag sequence. Untagged files fall back to raw drawing order — whatever sequence the generator happened to emit — which is why columns interleave and footers interrupt sentences.
- Artifacts — the skip-list. Running headers, footers, and page numbers must be marked so readers pass over them silently, instead of announcing "14" mid-paragraph.
- Alt text — descriptions attached to Figure tags. Meaningful ones carry the image's information ("bar chart: revenue rising 2020–2024"); an untagged file can't hold any.
- Metadata — language (so the right speech engine is used) and title (so tabs and readers say the document's name, not
final_v3.pdf).
How the layer gets added afterward
Remediation threads the structure into the existing file: content analyzed, roles assigned, brackets inserted between the original drawing operators, the tag tree and its indexes built, metadata repaired. Done properly, the original operators are untouched — Ceereo proves it by rendering every page before and after and comparing pixels, then re-running the machine checks on the result.
That's also the honest boundary: automation verifies and builds the machine-checkable layer. Whether a genuinely ambiguous magazine layout reads in the best possible order is a judgment call — flagged for human eyes, not guessed silently.
Related reading
Try Ceereo free
Upload a messy Word document and fix its formatting with AI, no signup required.
Check my PDF