How to tag a PDF for accessibility

Start with the job the reader needs to complete

When I tag a PDF, I start with a practical question: can someone who does not see the visual page structure understand the same document, move through it predictably, and reach its important information?

That question keeps the work focused. A PDF can look polished and still be difficult to use with assistive technology. Columns may be read in the wrong sequence. A heading may only be large, bold text. A chart may have no useful description. A table may sound like an undifferentiated stream of cells.

PDF tagging is the work of adding or repairing the document structure that communicates meaning beyond its appearance. The goal is not to make a checker report look clean. The goal is to make the actual reading experience clear and usable.

Understand what PDF tags do

Tags are structural labels stored in a PDF’s structure tree. They are not visible formatting applied to the page. A heading can remain visually identical after it is correctly tagged, but the tag tells software and assistive technology that the text is a heading and where it sits in the document hierarchy.

Common tags represent paragraphs, headings, lists, tables, figures, links, and sections. The Section 508 guide to common PDF tags is a useful reference for the roles these structures typically serve.

A meaningful tag tree lets a reader identify the document title, move among headings, understand a list as a list, and encounter a figure at the point where it is discussed. It also helps prevent page layout from becoming the only way to infer relationships.

Not every visible object needs a content tag. Decorative lines, purely ornamental images, and repeated page furniture can be artifacts when they add no information. Running headers, footers, and page numbers also often need artifacting so they do not interrupt every page with repetitive announcements. That decision depends on context: a page number referenced in the content may still be meaningful.

Start with the most usable source document

If the source file is available, I fix it before relying on PDF repair. Use real heading styles rather than manually formatted text. Build lists with list tools. Define table headers. Add image descriptions where the author knows the subject best. Set the document language and use link text that describes the destination or action.

Exporting from a structured source often produces a useful starting tag tree. It does not remove the need for review. Exports can flatten lists, mis-handle grouped objects, create empty tags, or produce a reading order that follows drawing order instead of the intended narrative.

Scanned PDFs require a separate approach. A scan is often an image of text rather than text itself. Run OCR when it is appropriate, then carefully correct recognition errors before building or repairing the structure. OCR can confuse characters, lose punctuation, merge columns, and misread table borders. If the recognized text is wrong, a well-organized tag tree cannot make the document understandable.

I treat automatic tagging the same way. It can speed up straightforward documents, but it is a draft. Complex page layouts, forms, sidebars, multi-column reports, figures, and tables need close human review.

For a broader remediation workflow, see How to remediate a PDF for accessibility. The tagging stage works best when it is planned alongside content corrections, metadata, and final review.

Inspect the structure tree before editing

Open the PDF’s tag or structure-tree view and compare the branches against each page. I look at the tree before making sweeping changes because it reveals whether the document has usable structure already or merely a collection of generic containers.

Look for these common failures:

• Missing content tags, especially on pages added late in production.
• Generic containers where headings, paragraphs, lists, or tables should be identified.
• Heading levels that skip or do not reflect the actual document hierarchy.
• Duplicate text caused by layered source content or OCR artifacts.
• Empty tags that create needless stops.
• Content placed under the wrong branch, such as a caption separated from its figure.

The structure tree should reflect the document’s intended hierarchy, not simply the left-to-right placement of objects on a page. A title followed by a first-level section heading and a subsection should be represented as a coherent structure, even if the visual design uses unusual spacing or typography.

Use the tree to assess semantics. Then use a reading-order view, reflow view where available, and direct assistive-technology testing to confirm the sequence exposed to users. These views can disagree, which is exactly why a single inspection method is not enough.

Apply the right tags to text and page structure

Tag ordinary body copy as paragraphs. Identify the document title and use headings for major sections and subsections. Keep the hierarchy logical. For example, do not make a visually prominent callout a top-level heading if it is not a new section of the document.

Use genuine list structures for ordered and unordered lists. Tagging each bullet as an isolated paragraph removes useful relationships, including list boundaries and item grouping. For a numbered procedure, the order is meaningful and should remain available to the reader.

Preserve emphasis when it changes meaning. A warning, a defined term, or a phrase that must be stressed may need semantic emphasis. But do not create structural noise merely because text has a different font, color, or size.

Artifact nonessential visual material. Repeated logos, decorative rules, background flourishes, and empty spacing objects should not create extra reading stops. Be careful with headers and footers: remove repetitive furniture from the reading experience, but retain content that contributes to orientation or meaning.

Tags are only part of the document setup. I also set a meaningful document title and the primary document language. When a passage changes language, that change may need to be identified as well. These properties affect how the document is announced and interpreted.

Repair reading order and multi-column layouts

Reading order is where an apparently well-tagged PDF can still fail readers. Confirm that content is encountered in a reasonable sequence: title, introductory material, section headings, body text, notes, captions, and referenced content.

Multi-column pages deserve slow review. Automated tools may read across both columns line by line, jump from a sidebar into the middle of a paragraph, or move a caption far away from the chart it explains. I check several points in each column, not only the first block of text, because errors often appear around tables, pull quotes, and page breaks.

Keep a caption associated with the figure or table it describes and place it at the appropriate point in the sequence. A caption that is visually close but announced much later creates confusion. The same applies to callouts, text boxes, sidebars, footnotes, and endnotes. Their visual position does not establish their semantic relationship.

Keyboard navigation is useful here as a practical test. If focus lands on links or interactive elements in an order that makes no sense, inspect both the page content and tag structure. For detailed troubleshooting patterns, see How to fix PDF accessibility issues.

Tag figures, links, and tables according to meaning

For an informative figure, write concise alternative text that communicates its purpose or the key information needed at that point in the document. Do not simply repeat “image of” or describe every visual detail. A useful description answers the question the figure is there to answer.

Some figures need more than short alternative text. A complex chart, process diagram, map, or technical illustration may require a longer explanation in nearby body text, a caption, or an associated description. Put the data trend, comparison, or conclusion into words. A short alternative-text field is rarely a good place for a full analysis.

Artifact decorative images and ornamental graphics so they are not announced as meaningful content. If an image contains text that matters, it is not decorative; the text and its purpose must be available in the reading experience.

Links should have meaningful text and be represented in the tag structure. Avoid vague link labels such as “click here” when several appear on the same page. The surrounding sentence can provide context, but the link itself should still be understandable when encountered on its own.

Tables need deliberate work. Build table semantics with a table, rows, header cells, and data cells. Identify headers and their relationship to the data, especially in tables with multiple header rows or row headers. The purpose is to preserve the question a header answers for each data cell.

Complex tables are a design issue as much as a tagging issue. If the table has merged cells, nested headings, irregular spans, or many cross-references, consider splitting it into smaller tables or explaining key results in prose. A simpler table is easier for everyone to interpret and easier to maintain accurately.

Validate with more than an automated checker

Run an accessibility checker to find likely issues, then investigate every result in the actual document. A green checker light is never a pass. Checkers can find missing titles, absent alternative-text fields, untagged content, or some table problems. They cannot reliably judge whether a description is useful, whether a heading hierarchy expresses the right meaning, or whether reading order makes sense to a person.

My final review includes the structure tree, heading hierarchy, reading sequence, figures, links, tables, document language, title, and artifacting. I also test keyboard navigation and, where possible, listen to the PDF with assistive technology. That is where duplicated content, unclear link announcements, misplaced notes, and confusing table traversal become obvious.

Record what was repaired, which review steps were performed, and any limitations that remain. If a scan was too poor to recover accurately or a source layout needs redesign, document that finding and feed it back to the next edition. A remediation workspace such as Ordecto can help teams inspect and repair tagged PDFs, reading order, figures, and tables while retaining evidence of the work completed.

The finished file should be judged by the reading experience, not by its appearance or one automated result. That mindset produces better tagged PDFs and makes future updates easier to handle.