Making a PDF accessible: what most documents get wrong

Accessibility 10 September 2026

A PDF page is a drawing, not a document

This is the part that is easy to miss: a standard PDF page does not inherently know what a heading is, what order its paragraphs go in, or that a column of numbers is a table. It knows where to draw a line of text and which font to use for it. To a sighted reader, layout on the page communicates all of that — bigger text near the top is obviously a heading. A screen reader has no access to that visual reasoning. Without extra information, it either reads the page in whatever raw internal order the text objects happen to be stored in, which frequently does not match the visual reading order at all, or it has almost nothing to read because the page is actually an image.

The extra information a PDF needs to be genuinely accessible is called tagging, and a PDF that has it is described as a Tagged PDF. Tags mark up the same structure HTML uses — headings, paragraphs, lists, tables, reading order — laid on top of the visual page. A PDF/UA-compliant document, a standard finalized in 2012 specifically for this, requires this tagging along with several other structural requirements. Most PDFs in circulation have none of it.

The most common ways a PDF fails at this

Scanned documents with no text layer. A photographed or scanned page is, to the computer, one large image. There is no text to read at all, tagged or not, unless OCR has been run to add a hidden text layer underneath the image. A screen reader gets nothing.

Reading order that does not match visual order. Multi-column layouts, sidebars, and pull-quotes are frequent offenders — a document exported from a page-layout tool can place text objects in the underlying file in whatever order the design software happened to create them, which has no relationship to the order a person would actually read the page.

Images with no alternative text. A chart, a diagram, a scanned signature — without a text description attached to the image object, a screen reader either skips it silently or announces nothing more useful than “image.”

Form fields with no labels. A fillable PDF form where the visual label “Date of Birth” sits next to a text box, but nothing in the file structure actually connects the two, leaves a screen reader user with an unlabeled blank field and no way to know what belongs in it.

Color as the only signal. A required field marked only in red, or a chart that distinguishes lines only by color, communicates nothing to someone who cannot see color differences, whether that is due to blindness, low vision, or color blindness.

What this actually requires, realistically

Doing this properly, for a complex document, generally means going back to the source file — the word processor or design file the PDF was exported from — rather than trying to fix it after the fact. Most modern word processors can tag headings, lists, and reading order automatically during export, provided the original document actually used real heading styles instead of just making text bigger and bold by hand. That single habit, using a Heading 1 style instead of manually formatting text to look like a heading, is responsible for a large share of whether the resulting PDF is accessible at all.

For a scanned document, the minimum fix is running OCR to add a searchable, readable text layer. It will not produce a fully tagged, reading-order-correct document, but a screen reader can at least access the words, which a raw scan does not allow at all.

Where a tool like Doxly fits, honestly

None of Doxly's tools currently add PDF/UA tagging or run OCR — building that well is a substantial undertaking, not a checkbox, and it would be dishonest to imply otherwise. What the tools here can do: page numbers, cropping, and organizing a document all preserve existing tags rather than stripping them, because they manipulate PDF objects directly instead of flattening the page to an image. Compression is the one exception worth calling out here directly — it works by rasterizing each page, which does strip tags and any text layer along with it. If accessibility matters for a document, compress it last, or not at all.