How to Fix Untagged PDFs So They Pass WCAG and Get Read by AI Search Tools

An untagged PDF fails WCAG because screen readers and AI crawlers can't tell a heading from a paragraph, or a table from a wall of text — they just see a flat image of characters. Fixing it means adding a tag structure (headings, lists, tables, reading order, alt text) so both assistive tech and AI models can parse the document the way they'd parse a well-built webpage.

I've audited a lot of sites where the HTML is clean, the color contrast passes, the forms are labeled — and then there's a folder of 200 PDFs sitting on the server from 2019, untouched since someone exported them from Word. Nobody thinks about the PDFs. That's usually where the audit falls apart.

What Makes a PDF "Untagged" (and Why It Fails WCAG)?

A PDF is untagged when it has no underlying structure map — no code layer telling a reader what's a heading, what's body text, what's a table cell, and what order to read things in. Visually, an untagged PDF can look perfect. Structurally, it's just a grid of text and image objects with coordinates.

WCAG 2.1 doesn't have a separate rulebook for PDFs. The same success criteria that apply to web pages apply here, because W3C's WAI guidance treats PDF as just another format that has to meet the same bar as HTML: 1.3.1 (Info and Relationships), 1.1.1 (Non-text Content, i.e. alt text), 2.4.6 (Headings and Labels), and 4.1.2 (Name, Role, Value) all apply directly. An untagged PDF almost always fails all four at once, because none of that information exists in the file.

Under the ADA, PDFs count as part of your "place of business" content if you're a public-facing company, and Section 508 makes tagging mandatory for federal agencies and contractors. So this isn't a gray area — it's one of the more clear-cut compliance gaps out there, and one of the most commonly missed. For the legal side of how ADA and WCAG relate, our ADA vs. WCAG breakdown covers what actually counts as a violation.

Why Do AI Search Tools Ignore Untagged PDFs?

AI crawlers skip or misread untagged PDFs because they rely on the same structural signals screen readers do — headings, semantic order, and extractable text — and an untagged PDF gives them none of it.

Here's the part most people don't expect: ChatGPT, Perplexity, and Google's AI Overviews don't "see" a PDF the way you do. They run text extraction on it. If your PDF was made by scanning a printed page (common with older government forms, product spec sheets, or old menus), there might be zero actual text in the file — just a picture of text. No OCR layer, no extraction, no citation. Ever.

Even when there's real text, an untagged PDF often extracts in the wrong order. Multi-column layouts are the classic failure: a two-column PDF export can pull text left-to-right across both columns instead of down one column and then the other, turning a coherent paragraph into scrambled nonsense. An AI model reading that isn't going to cite it — it can't even parse it into a sentence that makes sense.

This is the same underlying problem we talk about in our guide on what AI readability actually means: crawler access, a shared content core between what humans and bots see, and extractability. PDFs fail on all three if they're untagged, scanned without OCR, or built with a layout tool that doesn't preserve structure.

How Do You Check If a PDF Is Tagged?

The fastest check: open the PDF in Adobe Acrobat, go to the accessibility panel, and run "Full Check." It'll flag missing tags, missing alt text, and reading-order issues in about 20 seconds.

Don't have Acrobat Pro? A few other quick tests work almost as well:

• Try to select and copy text. If you can't select individual words — if the whole thing behaves like one image — it's a scanned document with no text layer at all.
• Check the document properties. In Acrobat or Preview, look for a "Tagged PDF" field under Advanced properties. It'll say Yes or No.
• Run it through PAC (PDF Accessibility Checker), a free tool built specifically for testing against PDF/UA and WCAG. It's the closest thing to an industry-standard free checker for this.
• Ask a screen reader to read it. NVDA (free, Windows) will either read the content in a sensible order or immediately expose the chaos — reading table cells in the wrong sequence, skipping images entirely, or announcing "no tags found."

A quick industry data point worth knowing: WebAIM's ongoing accessibility research consistently finds that PDF remediation issues rank among the most common — and most overlooked — failures on government and enterprise sites, right alongside missing alt text and poor color contrast on the HTML side. PDFs just don't get audited as often because they feel like "documents," not "website." They're both.

How Do You Fix an Untagged PDF?

You fix an untagged PDF by adding a tag tree that defines headings, paragraphs, lists, tables, and reading order — either by regenerating it correctly from the source file or by manually tagging it in Acrobat Pro.

Start at the source, not the PDF

If you have the original Word, InDesign, or Google Docs file, fix it there first. Use real heading styles (Heading 1, Heading 2), not bold 18pt text pretending to be a heading. Use the built-in list and table tools instead of tabs and spaces. Then export with tagging turned on — in Word, that's "Save as PDF" with the "Document structure tags for accessibility" box checked, not "Print to PDF," which strips everything.

This one habit — fixing the source instead of the export — fixes 80% of your accessibility backlog with almost no extra effort per document.

If you only have the PDF

1. Run OCR first if there's no selectable text. Acrobat's "Recognize Text" feature or a tool like ABBYY FineReader will add a text layer to scanned pages.
2. Add tags automatically, then check the results by hand. Acrobat's "Autotag Document" gets you 60-70% of the way there on a well-formatted file — worse on complex layouts, form-heavy PDFs, or anything with sidebars and pull quotes.
3. Fix the reading order using the Order panel. This is where multi-column layouts usually need manual correction.
4. Add alt text to every image, chart, and figure. Decorative images (borders, dividers) should be tagged as artifacts so screen readers skip them instead of announcing "image, unlabeled." For guidance on writing alt text that also works for AI extraction, see our alt text guide — the same principles apply inside a PDF as they do on a web page.
5. Tag tables properly with header rows (TH) distinct from data cells (TD), so a screen reader announces "Column: Price, Row: Widget A" instead of reading numbers with no context.
6. Set the document language and title in the file properties. Small step, easy to forget, required by WCAG 3.1.1.

What Belongs in an Accessible PDF Checklist?