PDF Accessibility
Accessible PDF from LaTeX: What to Fix Before You Publish
A LaTeX export is only the starting point for an accessible PDF. Learn what to check in tags, equations, reading order, tables, links, metadata, and screen-reader output before publishing.

An accessible PDF from LaTeX starts in the .tex source, but it is finished in the exported PDF. LaTeX can now produce structured tagged output in some workflows, yet export alone does not make every research paper, thesis, report, or proceedings submission ready for screen-reader users.
The practical job is to check the final PDF for tags, reading order, equations, tables, figures, links, bookmarks, metadata, and representative screen-reader output. If the document has complex math, multi-column pages, footnotes, or dense tables, assume it needs a PDF review after export.
Updated November 22, 2026.
What an accessible PDF from LaTeX requires
An accessible PDF must expose meaning, not just appearance. A sighted reader can infer that large bold text is a heading, a centered line is a title, or a numbered block is a list. Assistive technology needs that structure encoded.
A tagged PDF is a PDF with structural tags that identify the role and order of content: headings, paragraphs, lists, tables, links, figures, captions, and other objects. Visual formatting alone does not create those tags. A PDF can look polished and still be almost unusable if its underlying structure is missing or wrong.
For LaTeX users, this matters because the source file is often highly semantic in one sense: \section, \caption, \cite, and equation environments carry meaning. But the exported PDF must preserve that meaning in a form assistive technology can use.
The LaTeX Project has shown LaTeX-generated examples connected with WTPDF, PDF/UA-2, and PDF/A-4 work. As of November 2026, its PDF/UA-2 announcement says: “All of the examples have been produced by LaTeX using the code from our LaTeX tagged PDF project” (LaTeX Project). That is a strong sign of where the ecosystem is going. It is not a guarantee that every local template, package combination, journal class, or conversion workflow will produce an accessible final PDF.
PDF accessibility standards use several overlapping terms. PDF/UA is the ISO standard focused on accessible PDF structure. WCAG, the Web Content Accessibility Guidelines, is a broader set of digital accessibility criteria often referenced when reviewing PDFs. A checker can compare a file against parts of these models, but the final judgment still depends on whether the document’s content makes sense to a human reader using assistive technology.
If your source document is Microsoft Word rather than LaTeX, the export risks are different; see the separate workflow for an accessible PDF from Word. The LaTeX problem is more often about equations, custom classes, generated navigation, package interactions, and whether tagged output survived a specialized publishing pipeline.
Start with accessible structure in the LaTeX source
The cheapest accessibility fix is the one made before export. Once a PDF is generated, every missing heading, broken link, or ambiguous figure description becomes a remediation task.
Start with the document hierarchy. Use the proper sectioning commands for real structure, not just for typography. The title, abstract, sections, subsections, appendices, acknowledgments, and references should follow a logical order.
Do not skip levels for visual effect. A \subsection should not appear before the first \section, and a styled line should not stand in for a heading. Ohio State’s LaTeX accessibility guidance warns that header levels need to be adjusted before publishing when creating PDFs from LaTeX (Ohio State University).
Links need the same discipline. A visible URL may be useful in print, but it is often poor link text for a screen-reader user. Prefer descriptive text that tells the reader where the link goes or what it does.
For example:
Better: “Download the supplementary dataset”
Worse: “Click here”
Worse: “https://example.edu/data/final/v3/download?id=93847”
If a journal requires visible URLs in references, keep the reference format, but check whether the generated PDF exposes a useful link target and whether the link can be reached by keyboard.
Set the document title and language where your LaTeX workflow supports those metadata fields. Metadata is information about the PDF itself, such as its title and language. Screen readers and PDF readers may use it for document identification, pronunciation, and navigation.
Then inspect the exported PDF. Source declarations do not always survive the path through a class file, build system, post-processing step, or publisher template.
Figures need decisions before export too. Alt text is a text alternative for an image, used when the image cannot be seen. In a LaTeX document, some images are decorative, some are informative, and some are the argument of the paper.
Decorative rules, flourishes, or layout images should not be announced as meaningful content. Informative figures need a description that explains the point of the figure, not merely its appearance. “Scatterplot” is rarely enough; “Scatterplot showing a positive association between X and Y after excluding outliers” is closer to useful.
Captions help, but captions and alt text are not interchangeable. A caption may identify the figure; an image description may need to explain the visual evidence the reader would otherwise miss.
Generated bookmarks can improve navigation. Bookmarks are the clickable navigation entries shown by many PDF readers, usually derived from headings. They are useful, but they do not replace correctly tagged headings. A PDF can have good bookmarks and still have broken tags or a bad reading order.
LaTeX features that commonly need a closer PDF review
The risky parts of a LaTeX PDF are usually the parts that made LaTeX attractive in the first place: math, references, complex layouts, tables, and custom publication formats.

Equations
Mathematical content is the highest-friction area in many LaTeX PDFs. A screen reader cannot infer the meaning of an equation from pixels. The PDF needs an equivalent representation, such as MathML, spoken text, or another form that assistive technology can interpret.
MathML is a markup language for representing mathematical notation in a machine-readable way. For PDFs, it can help preserve structure that is lost when formulas are treated as visual objects. The PDF Association has described newer PDF mechanisms that allow MathML and LaTeX representations to be embedded as supplements to the visual math content (PDF Association).
That still does not make math automatic. Generated MathML or spoken text may be wrong, too verbose, too sparse, or missing the context a reader needs. A person who understands the subject should review important formulas, especially in STEM papers, financial reports, statistical appendices, and technical standards.
For a deeper explanation of the format problem, see Ordecto’s guide to accessible math in PDF and its article on why equations change the scope of PDF accessibility remediation.
Tables
Tables need more than grid lines. A screen reader needs to know which cells are headers, which cells are data, and how the relationships work.
Simple tables may export cleanly. Complex tables often do not. Watch for:
multi-row or multi-column headers
merged cells
nested tables
tables used for visual layout
footnotes inside tables
mathematical notation inside cells
rotated labels
repeated headers across page breaks
A table that is visually clear may be structurally ambiguous. If a reader cannot tell which header applies to a data cell, the table needs repair or redesign.
Multi-column layouts
Many LaTeX papers use two-column formats. That is common in conference proceedings and journal submissions, but it creates a reading-order risk.
Reading order is the sequence in which the PDF content is exposed to assistive technology. It should match the order a reader is meant to consume the document.
Bad reading order can make a PDF sound like nonsense. A screen reader may read down the left column, jump to a sidebar, move to a figure caption, return to the title, and then continue in the right column. On the page, everything looks correct. In the tag tree, the argument has been scrambled.
Two-column layouts, marginal notes, pull quotes, sidebars, and floats all deserve manual review.
Footnotes and bibliographies
Footnotes create a different problem: interruption. The reader needs to know that a note exists, reach it when appropriate, and return to the main text without losing the thread.
Check footnote markers, note content, and any return links. The note should not appear unpredictably in the middle of an unrelated sentence. Bibliographies also need review because generated reference lists can contain links, punctuation, italics, small caps, URLs, DOIs, and line breaks that confuse reading order.
Figures and captions
Every figure needs a decision: informative, decorative, or redundant.
An informative figure needs a useful description. A decorative image should be marked so it is skipped by assistive technology. A redundant image may be skipped if the same information is fully available in surrounding text.
Captions should be associated with the correct figures. This sounds obvious until a PDF has floating figures, two-column layouts, or a template that places captions far from the image in the underlying structure.
Hyperlinks, bookmarks, and generated navigation
Review visible link text, link targets, keyboard access, and bookmark labels. A bookmark tree should reflect the real section structure, not a flat list of visual headings.
Generated navigation is helpful only when it matches the document. If the table of contents, bookmarks, headings, and tags disagree, screen-reader users get competing maps of the same file.
The non-obvious failure mode is this: a visually polished research paper can still be hard to use because its tags, equations, columns, or footnotes are in the wrong logical order. Accessibility is not a surface finish.
A post-export workflow for checking the PDF
Once LaTeX has produced the PDF, review the PDF itself. Do not stop at the .tex file.
The exported file is what the reader receives, what the journal posts, what the university archives, and what the public body publishes. That is the object that needs inspection.
1. Confirm document properties
Open the final PDF and check the basics:
document title
document language
author and subject metadata, if used
bookmarks
page labels
whether the file is tagged at all
Harvard’s PDF accessibility guidance lists checking tags, reviewing reading order and tags, ensuring image accessibility, and setting file properties as core PDF accessibility practices (Harvard Digital Accessibility Services).
A missing title or language is usually easy to fix compared with a broken tag tree, but it still matters. It also signals whether the export workflow is preserving accessibility-related information.
2. Inspect tags and structure
Check whether the PDF contains structural tags for headings, paragraphs, lists, tables, links, figures, and artifacts.
An artifact is content that should be ignored by assistive technology, such as decorative lines, page backgrounds, or repeated visual ornaments. Page numbers and running headers may need careful handling depending on whether they help or interrupt navigation.
Look for these problems:
headings tagged as plain paragraphs
paragraphs split into many fragments
lists not tagged as lists
table cells not associated with headers
figures with no useful description
decorative content announced as real content
links with unclear purpose
tags ordered differently from the intended reading sequence
Do not assume that a tagged PDF is correctly tagged. Tags are necessary, not sufficient.
3. Test reading order on representative pages
Do not review only page one. Pick pages that stress the layout.
For a LaTeX document, representative pages usually include:
title page
abstract
table of contents, if present
dense equation page
two-column page
page with a figure and caption
page with a complex table
page with footnotes
bibliography page
appendix page
any page likely to be shared or read independently
Read the tags in order, or use a PDF tool that exposes reading order. The question is simple: if this sequence were spoken aloud, would the argument still make sense?
4. Review figures and formulas with a person
Automation can detect missing structure. It cannot reliably decide whether a figure description captures the scientific point or whether a formula’s spoken version preserves the intended notation.
A person should review:
equations central to the argument
symbols that depend on domain convention
matrices, aligned equations, fractions, and nested expressions
charts where trend, outliers, or confidence intervals matter
diagrams with arrows, spatial relationships, or process flow
figures where color or shape encodes meaning
For math-heavy publication, Harvard’s accessible math guidance recommends a “Plus-One” approach: keep PDFs for printing and add a second format so equations become more navigable (Harvard Digital Accessibility Services). That can be a pragmatic choice when the PDF is required but not the best reading format for every user.
5. Check tables and links
For each important table, verify whether headers and data cells are exposed correctly. If the table is complex, ask whether it should be simplified, split, or accompanied by a text explanation.
For links, check:
visible text describes the destination or action
destination works
link is keyboard reachable
link annotation is tagged
references, DOIs, and URLs do not create noisy reading
Academic PDFs often contain dozens or hundreds of links. Broken or undescriptive links become more painful at that scale.
6. Run a PDF/UA- or WCAG-oriented checker
Use an automated checker to find technical issues and record remediation evidence. Treat the output as a findings list, not as proof that the PDF is accessible.
A checker may identify missing tags, missing titles, untagged annotations, table structure problems, or figures without descriptions. It may also flag items that need a person because intent cannot be inferred safely.
Automated checking is useful because it is systematic. It is limited because accessibility depends on meaning. Ordecto’s article on what automated PDF accessibility checkers cannot test covers that boundary in more detail.
7. Test representative pages with a screen reader and keyboard
A screen-reader spot check should complement the structural review. It should not replace it.
Test the same representative pages selected for reading order. Listen for:
document title announcement
heading navigation
list announcement
table navigation
link purpose
figure descriptions
equation output
footnote behavior
bibliography readability
unexpected jumps in reading order
Keyboard testing matters too. If links, form fields, or navigation elements exist, a mouse-only workflow is not enough.
When to use a PDF remediation tool after LaTeX export
Use a remediation tool when the exported file needs structured checking and repair rather than another source edit. That often happens when the source is stable, the deadline is close, the publisher requires PDF, or the document has already gone through a template or production system that you cannot easily change.
At that point, Ordecto, a PDF accessibility tool from the team behind Otio, can check the exported PDF against PDF/UA-1 with WCAG 2.2 as a referenced test model. Its findings are grouped as failed, needs a person, or not tested, rather than presenting a machine result as proof that the file is accessible.
A useful handoff looks like this:
First, run the exported file through a free PDF accessibility checker to identify structural issues in the actual PDF.
If the tag tree or reading order needs repair, use a PDF remediation workflow to repair headings, paragraphs, lists, tables, links, figures, decorative content, and reading order.
For a standards-oriented review, use a PDF/UA checker and keep the findings with the publication record.
The important constraint remains: reading order, the meaning of figures, complex tables, and mathematical correctness still need human judgment. A repaired PDF should be treated as a candidate for review, not a certificate.
For LaTeX documents, that distinction matters most with equations and dense technical layouts. A remediation tool can repair structure and produce a report describing what was checked, what changed, and what still needs a person. It cannot decide, alone, whether a formula’s accessible representation is mathematically sufficient for every reader.
Final checks before publishing the LaTeX PDF
Before publication, review the final PDF, not an earlier draft. Accessibility can break late: after a bibliography rebuild, class-file change, figure replacement, publisher conversion, or PDF optimization step.
Use a representative-page review rather than checking only the first few pages. Include the hardest content in the document:
most complex equation
most complex table
first multi-column page
page with a figure and caption
page with footnotes
bibliography
appendix
any standalone handout or supplement page
Confirm that the final PDF contains the intended title, language, tags, links, bookmarks, figure descriptions, equations, table structure, and reading order. If the document is going to a journal, institutional repository, course site, public agency page, or conference system, keep the final accessible version separate from the draft export.
Preserve the checker report, manual review notes, and repaired file with the publication record. That gives the team a defensible account of what was tested, what was changed, and what required human judgment.
The next practical step: upload the exported LaTeX PDF to Ordecto’s free PDF accessibility checker to see which structural issues need repair and which findings still need human review.
FAQ
Q: Can LaTeX automatically create an accessible PDF?
A: LaTeX workflows can produce tagged PDF output, but no engine or export option automatically makes every document accessible. The exported PDF still needs checks for structure, reading order, equations, tables, figures, links, metadata, and screen-reader behavior.
Q: Are tagged PDFs and accessible PDFs the same thing?
A: No. A tagged PDF exposes structural roles such as headings, paragraphs, lists, tables, and figures, which is necessary for many assistive-technology workflows. Tags alone do not prove that reading order, descriptions, equations, complex tables, or navigation are correct.
Q: How should equations in a LaTeX PDF be reviewed?
A: Check whether equations have an equivalent representation such as MathML or spoken text, then have a person verify the mathematical meaning and correctness. Automated output can miss notation, relationships, or context that a reader needs.
Q: Does a PDF/UA or WCAG checker prove that a LaTeX PDF is accessible?
A: No. An automated checker can identify many technical findings, but it cannot reliably judge intended reading order, the meaning of every figure, or the correctness of complex tables and formulas. Combine automated results with manual and representative screen-reader testing.




