PDF Accessibility
Accessible PDF from Word: What to Check After Export
Exporting an accessible Word document does not guarantee an accessible PDF. Use this post-export workflow to check tags, reading order, images, tables, metadata, forms, text layers, and human-review evidence before release.

If you searched for “accessible PDF from Word,” the key point is this: a clean Word accessibility check and the right export settings are only the start. The exported PDF has its own tags, reading order, metadata, text layer, links, forms, and tables, and those must be reviewed before the file is released.
A tag is structural information inside a PDF that identifies headings, paragraphs, lists, tables, links, figures, form fields, and other content. Screen readers and other assistive technology use those tags, plus the PDF’s reading sequence and metadata, rather than the Word document you started from.
Use a simple release workflow: export from Word, inspect the PDF structure, repair what changed or broke, manually review the hard cases, document the result, and only then publish.
Updated November 16, 2026.
An accessible PDF from Word still needs its own review
Microsoft’s own guidance says the fastest way to help ensure a PDF is accessible is to run Word’s Accessibility Checker before generating or saving the file as a PDF. That is good preparation, not proof that the PDF itself is usable. The exported file is a new technical object, and it can lose or distort information that looked right in Word.
University accessibility teams tend to describe the same sequence: make the Word source accessible, export it correctly, then review the new PDF. James Madison University’s guide, for example, separates the job into reviewing the source document, exporting the Word document as a PDF, and reviewing the accessibility of the new PDF (JMU).
That final review matters because a PDF viewer and a screen reader do not read “your Word file.” They read the PDF’s own:
Tag tree: the structure that identifies headings, paragraphs, lists, tables, figures, links, and form fields.
Reading order: the sequence in which assistive technology encounters page content.
Text layer: selectable, machine-readable text behind the visual page.
Metadata: document language, title, and related properties.
Interactive structure: form field names, keyboard focus order, link targets, and controls.

The practical order is:
Export from Word using accessibility-aware PDF options.
Inspect the PDF’s structure, not just its appearance.
Repair broken tags, reading order, missing text, link structure, forms, and metadata.
Manually review the parts a machine cannot understand.
Keep evidence of what was checked and changed.
Publish the reviewed final PDF, not an earlier draft.
Start with the PDF structure Word exported
Open the exported PDF and check whether it has a logical tag tree. The University of North Carolina’s Digital Accessibility Office calls the tag tree the most important element of PDF structure because it defines headings, lists, tables, images, and hierarchy for assistive technology (UNC).
Start with the broad structure:
One meaningful top-level heading: The document should have a clear main heading, not several competing first-level headings created by visual formatting.
Sensible heading hierarchy: Headings should move in a logical order. A report title followed by major sections, subsections, and sub-subsections should be represented as structure, not just bold or large text.
Paragraphs tagged as paragraphs: Body text should not be broken into stray spans, empty objects, or unrelated fragments.
Lists tagged as lists: Bullets and numbered items should be identified as lists, with list items in the correct order.
Figures tagged as figures: Informative images should be identifiable as figures, while decorative objects should not interrupt the reading sequence.
For a deeper explanation of the underlying structure, see Ordecto’s explainer on what a tagged PDF is. The short version for a Word export is this: visual styling is not enough. A heading that only looks like a heading may not behave like one in the PDF.
Next, check reading order. Reading order is the sequence in which assistive technology encounters the content on the page. A sighted reader can scan a two-column page, skip a repeated footer, notice a caption belongs to a chart, and ignore decorative lines. A screen reader depends on the PDF structure to make that sequence coherent.
Pay special attention to content that often moves out of order after export:
Multi-column text.
Text boxes.
Pull quotes and sidebars.
Captions.
Footnotes and endnotes.
Headers and footers.
Floating images.
Tables split across pages.
Watermarks and decorative separators.
A common failure is that the PDF reads a left column, then a footer, then the right column, then a caption from a previous page. Another is that a sidebar interrupts the middle of a sentence. These problems may not show up in Word’s document checker because they occur in the exported PDF’s structure.
For longer reports, also review bookmarks. Bookmarks are navigational links in the PDF sidebar, usually generated from headings. They should reflect the document’s main structure without duplicating, skipping, or flattening important sections. A 70-page report with no useful bookmarks may still contain headings, but navigation will be harder for many readers.
Review images, links, tables, and document metadata
Alt text is a short description of an informative image that communicates its relevant meaning. The key word is meaning, not appearance. Section508.gov defines alternative text as descriptive text that conveys the meaning of an image in digital content (Section508.gov).
Do not assume alt text copied from Word is still correct in the PDF. It may be missing, attached to the wrong object, too vague, or no longer appropriate after layout changes. A person needs to decide whether the description fits the image’s purpose in this document.
For images, ask:
Is the image informative or decorative?
If informative, does the description explain what the reader needs to know?
If decorative, is it excluded from the meaningful reading sequence?
If it is a chart, does the text convey the conclusion or data pattern, not just “bar chart”?
If the same image appears repeatedly, should it be described once and then treated as decorative elsewhere?
Harvard’s digital accessibility guidance makes the same point in plain terms: alt text should provide context on how the image relates to the surrounding content, not merely describe the image in isolation (Harvard). For remediation details, use Ordecto’s guide on how to add alternate text in PDF.
Links need a separate check. Link text should make sense out of context because assistive technology users may navigate through a list of links rather than reading every paragraph around them. “Read more,” “here,” and long raw URLs are usually poor labels unless the surrounding structure makes their purpose clear.
Check both the label and the destination. Repeated links should not create confusing navigation. If five identical “download PDF” links point to five different files, the labels need more context.
Tables are usually where Word exports look better than they behave. A simple table with one header row may export cleanly. A table with merged cells, nested tables, multi-level headers, blank layout cells, or headers that span columns often needs human review.
For each table, check:
Whether the header row or header column is identified as a header.
Whether data cells are associated with the right headers.
Whether the reading sequence preserves the meaning of each cell.
Whether merged or spanning cells make sense when read linearly.
Whether the “table” is really being used for page layout rather than data.
Document metadata is less visible, but still important. Check the PDF’s document language and title. The title exposed to assistive technology should be meaningful, such as “Annual Safety Report 2026,” not a filename like final_report_v7_exported.pdf.
If the PDF title is missing or wrong, users may hear an unhelpful file name in the viewer or screen reader. If the language is missing, assistive technology may use the wrong pronunciation rules.
Check text, forms, equations, and keyboard access
A PDF that looks like text may still be an image. Scanned pages, photographed forms, and some exported files contain visual text without a usable text layer. A text layer is the machine-readable text behind the visual page; without it, a screen reader may have little or nothing to read.
NCSU Libraries explains optical character recognition, or OCR, as the process of detecting text in an image and embedding it in the document so the text becomes searchable (NCSU Libraries). OCR is necessary for image-only pages, but it is not the end of accessibility work. Recognized text still needs correct reading order, headings, tables, and image treatment.
Do a quick test: try selecting a sentence with the cursor. If the entire page selects as one image, or the selected text is nonsense, the PDF needs OCR and review. If the text is selectable but appears in the wrong order when copied, the reading order may still be broken.
Forms need their own inspection. If the Word document became a fillable PDF, check:
Field names that describe the purpose of each field.
Visible labels associated with the correct fields.
Keyboard focus order that follows the intended form sequence.
Instructions that are available to assistive technology.
Error messages or required-field indicators that are not color-only.
Buttons that are named according to what they do.
Keyboard access is the fastest way to expose many form and interactive failures. Move through the PDF using the keyboard, not the mouse. Check whether focus reaches headings, links, fields, buttons, and other controls in an expected sequence. Record where focus disappears, skips content, lands on decorative material, or jumps to an unrelated part of the page.
Equations need special attention. A math expression may look perfect on the page and still be useless when spoken as disconnected symbols. For equations, confirm that the spoken or text alternative communicates the mathematical meaning rather than merely describing the visual layout.
Visual inspection alone cannot catch these issues. An automated checker can identify many structural failures, but keyboard and assistive-technology review exposes the experience: what a reader hears, where focus moves, and whether the structure actually supports understanding.
Use automated checking to find failures, not to declare the PDF finished
After export, use an automated checker to find machine-detectable problems and organize the review. Ordecto, a PDF accessibility tool from the team behind Otio, offers a free PDF accessibility checker that checks each page against PDF/UA-1 with WCAG 2.2 as a referenced test model and categorizes findings as failed, needs a person, or not tested.
PDF/UA is the PDF accessibility standard for universally accessible PDF documents. WCAG means the Web Content Accessibility Guidelines, the web accessibility guidelines often referenced by digital accessibility policies. Referencing PDF/UA or WCAG in a tool does not certify that a PDF conforms to either standard.
That distinction matters. A checker can tell you that a heading tag is missing, a figure lacks a description, a scanned page needs text, or a form field has no accessible name. It cannot reliably decide whether a chart description communicates the right analytical point, whether a complex table makes sense, or whether a sidebar belongs before or after the paragraph it visually interrupts.
Ordecto’s repair workflow can tag headings, paragraphs, lists, tables, links, figures with descriptions, form fields, and decorative artifacts; repair reading order; and create a text layer for scanned pages. Its PDF remediation page explains the repair side, while the PDF/UA checker is the better fit when the question is standards-oriented findings.
Use the checker result as a work queue:
Failed: repair the structural issue or decide that the source PDF needs to be rebuilt.
Needs a person: inspect the page, object, or sequence and record the judgment.
Not tested: understand that the absence of a machine finding is not evidence that the item is correct.
Section508.gov’s PDF testing training makes the same operational point: before using Acrobat’s accessibility check tools, the PDF must be visually verified to identify issues automated testing cannot detect (Section508.gov). In practice, the order is not “checker, then done.” It is “checker, repair, human review, rerun, document.”
A machine result is not a certificate. A repaired file is a candidate for review and release evidence, not proof of PDF/UA or WCAG conformance.
Complete a short human review before publishing
The human review does not need to become a full forensic audit for every short handout. It does need to cover the failure modes Word exports commonly leave behind.
Use a focused sequence:
Open the PDF and navigate without a mouse.
Move through headings and links.
Read several representative pages with a screen reader or equivalent assistive-technology check.
Inspect at least one typical table.
Review informative images and charts.
Test form fields, if the PDF has forms.
Check equations, footnotes, multi-column pages, and complex layouts where they appear.
Rerun the automated check after repairs.
Do not check only the first page. Word exports often look fine on the cover and fail later in the report, where tables, charts, columns, and footnotes appear.
Pick representative pages:
The title page.
A dense text page.
A page with headings and subheadings.
A multi-column page.
A page with a table.
A page with an image, chart, or diagram.
Any page with a form field.
Any page with equations or technical notation.
A page with footnotes or endnotes.
During the review, compare the assistive-technology experience with the intended visual meaning. A tag can be present and still wrong. A heading can be tagged but at the wrong level. A table can be tagged but impossible to understand. An image can have alt text that is grammatically polished and still misses the point.
For the deeper boundary between automated and human testing, see Ordecto’s article on what automated PDF accessibility checkers cannot test.
Record findings by page or object. A useful note is specific: “Page 14, revenue table: column headers not associated with data cells; repaired and rechecked,” or “Page 22, process diagram: description reviewed by report owner and accepted.” Avoid vague notes like “accessibility checked.”
Keep release evidence with the final PDF
Keep the materials that prove which file was reviewed and what happened to it. At minimum, retain:
The source Word document.
The first exported PDF.
Any repaired PDF output.
Human-review notes.
The automated accessibility report.
The final PDF that was actually published.
Use versioned filenames or another traceable record. The report should clearly identify the PDF it belongs to. If the final published file is policy-report-2026-final.pdf, the evidence should not point to policy-report-draft3.pdf.
The report documents what was checked, what changed, and what still needed a person. It is evidence of a workflow, not a certificate of PDF/UA or WCAG conformance.
This becomes especially important when a Word file changes after review. If someone edits the Word document, exports a new PDF, and overwrites the published file, the old evidence no longer applies. Repeat the post-export review for the new PDF, even if the edits look small.
The next action is simple: upload the exported PDF to the free PDF accessibility checker to identify machine-detectable issues, findings that need a person’s review, and repairs to make before release.
FAQ
Q: Can an automated checker decide whether a PDF’s alt text is correct?
A: It can identify missing or structurally invalid image descriptions, but it cannot reliably decide whether a description communicates the image’s meaning in context. A person must review informative images and complex charts.
Q: What should I do if the exported PDF is scanned or image-only?
A: Confirm that it has a usable text layer and that the recognized text is accurate. OCR can make words available to assistive technology, but the resulting reading order, headings, tables, and image meaning still require review.
Q: Does a PDF accessibility report prove conformance?
A: No. A report records automated findings and review decisions, but a tool result is not a certificate of PDF/UA or WCAG conformance. Confirm any formal obligations with your accessibility office or legal counsel.




