Accessible PDFs: What Tags, Reading Order, and Alt Text Actually Do
A sighted reader glances at a page and instantly knows that the large bold line is a heading, that two columns should be read left before right, and that the chart carries the argument. None of that is obvious to software. Visually, a PDF is just marks positioned on a canvas, with no inherent statement about what any mark means. Assistive technology needs that meaning stated explicitly, and the mechanism for stating it is the tag tree, an invisible structural outline stored alongside the visible page.
Tags label each piece of content with its semantic role: heading level one, paragraph, list item, table header cell, figure, link. A screen reader uses these labels to let users navigate the way sighted readers skim, jumping heading to heading or listing every table on the page. Without tags, the reader falls back to guessing from raw position data, and complex layouts collapse into nonsense. This is also why exporting from a word processor with proper heading styles produces a far more accessible PDF than one where headings were faked with bold text.
Reading order is a separate property, and it is where most documents quietly fail. The order in which content sits in the file is often the order the software drew it, not the order a human would read it. A newsletter can render perfectly while its underlying order interleaves the two columns line by line, producing an audio stream that switches topics every few words. Sidebars, pull quotes, headers, and footers can also be read in the middle of a sentence unless they are marked as artifacts and skipped.
Alternative text covers everything visual. Each meaningful image, chart, or diagram needs a short description conveying its purpose rather than its appearance. A revenue chart should not be described as a bar chart with blue bars, but as revenue rising from 2.1 to 3.4 million between 2023 and 2025. Purely decorative graphics should be marked as artifacts so they are ignored entirely, because announcing every divider line and background flourish is noise that makes documents exhausting to hear.
Before any of this matters, the document needs real text. A scanned page has no characters at all, so a screen reader announces an empty page regardless of how carefully it was laid out. Running OCR is therefore step one: it recognises the shapes and adds a genuine text layer, which also makes the file searchable and copyable. PdfWill runs OCR locally in your browser for English and Chinese and can output a searchable PDF or plain text. Full tagging and reading-order repair still need a dedicated accessibility editor, but a text layer is the foundation.