Guide

How to handle PDFs on a website you are making accessible

A site audit that ignores its documents has audited the wrapper. Which PDFs are in scope, how to find them all, and the order to work through them.

About 13 minutes · moderate

The part of the audit that gets skipped

An organisation runs an accessibility audit. The findings come back about the templates, the navigation, the colour palette, the forms. The audit is delivered, the fixes are scheduled, and the site improves.

Meanwhile the site serves nine hundred PDFs, and every one of them is content the audit did not look at.

This is the normal shape of the problem, and the reason is not negligence. Automated site crawlers report on HTML. Documents are somebody else’s department. And the number, once counted, is frightening enough that counting it feels like the wrong move.

Count it anyway. You cannot plan around a number you do not have.

Step 1 — Find them all

Three sources, and you need all three.

Crawl the site. Any crawler that follows links and records file types. wget --spider -r works; so does a commercial crawler you may already have.

wget --spider -r -l inf --no-verbose --reject-regex '.*' https://example.org 2>&1 \
  | grep -oE 'https?://[^ ]+\.pdf' | sort -u > site-pdfs.txt

List the CMS media library. The crawl finds linked documents. It does not find documents that are uploaded and linked from a page that is no longer published, or linked from an email, or simply known by URL — and those are still served.

Check the file system or bucket. Anything sitting in a directory the web server exposes is published, whether or not anything links to it.

Deduplicate. Expect the total to be two to five times what anyone estimated.

Step 2 — Classify before you remediate

Do not start at the top of the list. Four buckets, and the sorting is the highest-value hour in the project.

Should not be a PDF at all. A one-page notice, a contact list, an events calendar, a form people are meant to fill in. Converting these to HTML is faster than remediating them, produces a better result, and removes them from the problem permanently. This bucket is usually larger than people expect and it is where the biggest win is.

Must stay a PDF and is current. Documents of record: reports, statutory notices, signed documents, anything cited or printed. These need remediating and they are the real work.

Superseded. Old versions still being served alongside the current one. Unpublish them. A document that should not be available is not a remediation task.

Genuinely archived. Not used, not updated, kept for the record. Several instruments treat archived content differently — confirm your own position with counsel, and if it is out of scope, label it clearly rather than leaving it ambiguous.

Step 3 — Order the work by traffic, not by date

Pull page views per document from your analytics. The distribution is always steeper than anyone predicts: in most organisations a few dozen documents account for the large majority of downloads.

Remediate in traffic order. A hundred documents into a nine-hundred-document backlog, you will have fixed most of what anyone actually opens — which is both the right outcome for readers and the one that is defensible if you are asked about progress.

The triage guide covers how to run the queue itself.

Step 4 — The two bulk operations, first

Before remediating anything individually, run two things across the whole in-scope set. Both are metadata-only, both are safe to batch, and together they improve every document measurably.

Set the document title and the display flag. The most common accessibility failure in existence, fixed in a loop. See the title guide.

Set the document language. One field, and it decides which synthesiser voice reads the file. For Hebrew or Arabic documents this is the difference between usable and silent.

Neither touches page content, so neither can change how anything looks, and neither needs review.

Step 5 — Fix the links while you are there

The document is only half of it. The link on the page matters too:

• Say what it is, not “click here” or the filename. See links and annotations.
• Say the format and size: “Annual Report 2025 (PDF, 2.4 MB)”.
• Do not open a new window without saying so. target="_blank" with no warning is disorienting for a screen-reader user and worse on mobile.
• Give the alternative if there is one. If an HTML version exists, link both, and put the HTML first.

Step 6 — Stop the inflow

Everything above is remediation of a backlog. Unless the pipeline changes, next year’s backlog is the same size.

Three changes, in descending order of effect:

Fix the templates. Most organisations produce documents from a handful of Word or InDesign templates. Making those templates produce correct headings, real lists, tagged tables and a title field is a one-time change that improves every document made from them afterwards. This is the single highest-leverage action available.

Add a publishing gate. No document goes on the site without passing a validator and a two-keystroke paste test. A gate that runs in seconds gets used; one that requires a specialist does not.

Train the authors, briefly and specifically. Not a general accessibility course. Twenty minutes on: use heading styles, use the list buttons, set the header row, write alt text, set the title. Five habits, each of which prevents a category of failure downstream.

What to write down

For each document: the URL, whether it is in scope, its bucket, its traffic, what was done to it, when, by whom, and what remains. This is not bureaucracy — it is the answer to the question you will be asked, which is never “is the site accessible” but “what has been done about the documents, and what is left”.

A spreadsheet is sufficient. A folder of remediated PDFs with no record is not, because six months from now nobody will be able to say which of them were checked and which were merely re-saved.

Related failure conditions

• 01-005 Content is neither marked as Artifact nor tagged as real content Every text-showing, image-painting and path-painting operator on a page must sit either inside a marked-content sequence that maps to the structure tree, or inside an artifact sequence. Content that is in neither is orphaned.
• 06-001 Document does not contain an XMP metadata stream The catalog has no /Metadata entry pointing at an XMP packet. PDF/UA-1 requires one — it is where the conformance claim and the document title live.
• 09-001 Tags are not in logical reading order The order of elements in the structure tree does not match the order a human would read the page. This is the single most common substantive defect in tagged PDFs and the one Acrobat's own tooling handles worst.
• 11-001 Natural language for text in page content cannot be determined No /Lang is in effect for some text: neither on the catalog, nor on an enclosing structure element, nor on a marked-content sequence.

Last updated 2026-09-10 .