A PDF page is a set of drawing instructions with positions. Nothing in it says which text comes first — the visual order a sighted reader infers from layout is not recorded anywhere in the page content. Reading order is the tagged structure that supplies it, and getting it wrong is the single most consequential accessibility failure in PDF, because a document with perfect tags in the wrong sequence reads as fluent nonsense.
Three different orders in one file
The confusion here is that a PDF has several orderings that can all disagree.
Content stream order — the sequence in which drawing operations appear in the page. This follows whatever the producing application emitted, which may be by text frame, by layer, or by nothing coherent at all. Copy-and-paste and naive text extraction follow this order, which is why copying from a two-column PDF often gives you interleaved columns.
Visual order — how a sighted reader traverses the page: left to right, top to bottom, respecting columns, skipping the sidebar until the end. This exists only in the reader's head.
Structure order — the sequence of elements in the tagged structure tree. This is what a screen reader follows, and it is the only one you can control directly.
The job is to make structure order match visual order. When it does, the document reads correctly aloud, exports cleanly to other formats, and reflows sensibly on a small screen. When it does not, the reader gets the page number in the middle of a sentence and a caption before the paragraph it belongs to.
What good reading order looks like
Some specifics that recur in real documents:
- Headings come before the content they introduce, at the right level, with no level skipped.
- Multi-column text runs down column one, then down column two — never across.
- Captions follow their figure (or precede it, consistently), rather than appearing three paragraphs away.
- Pull quotes and sidebars come at a sensible break, not mid-sentence in the main flow.
- Page headers, footers, page numbers and decorative rules are artifacts — excluded from the reading order entirely, so they are not announced on every page.
- Tables are traversed as tables, cell by cell within rows, with header associations intact — see making accessible PDF tables.
- Lists are single list structures rather than a series of unrelated paragraphs, so the reader is told "list of six items".
- Footnote references and footnote text are linked so the reader can follow and return.
Inspecting the current order
Acrobat's Order panel. View → Show/Hide → Navigation Panes → Order overlays numbered boxes on the page showing the sequence. This is the fastest visual check available and it should be the first thing you look at on any document you did not produce. Drag entries in the panel to reorder.
Acrobat's Tags panel. The authoritative view. It shows the structure tree itself — the parent-child relationships, the element types, and the actual order the reader will follow. The Order panel is a simplification; the Tags panel is the truth.
Read Out Loud. View → Read Out Loud → Read This Page Only. Crude, and instantly informative. Thirty seconds of listening to a page reveals problems that visual inspection misses entirely.
PAC (PDF Accessibility Checker), free on Windows, has a structure preview that renders the document as a screen reader would encounter it. This is the closest thing to a ground-truth check without a screen reader.
An actual screen reader — NVDA on Windows (free), VoiceOver on macOS. Reading two or three pages of your own document with a screen reader is uncomfortable, revealing, and worth doing at least once for anyone responsible for accessible documents.
Fixing order in Acrobat
Two levels of intervention.
Reordering with the Order panel works when the elements are correctly tagged but in the wrong sequence. Drag items into place; the structure tree updates. Adequate for a page or two, tedious for a report.
Editing the Tags panel is required when the tagging itself is wrong — a heading tagged as a paragraph, a list flattened into loose text, two columns merged into one element, a caption absorbed into a figure. You retag by changing element types, dragging nodes to new parents, splitting and merging elements. It is fiddly work and there is no shortcut.
Marking artifacts: select the header/footer/decoration in the Content panel and choose Create Artifact, or use Tools → Accessibility → Reading Order to select a region and mark it as background. Getting the running heads out of the flow is usually the single biggest improvement to how a long document reads aloud.
The Reading Order tool (Tools → Accessibility → Reading Order) lets you draw boxes on the page and assign types — Text, Heading 1, Figure, Table, Background. It is the most approachable of the three and the least precise; it is a good first pass and rarely a sufficient last one.
The economics: fix the source, not the PDF
Remediating reading order in Acrobat takes anywhere from twenty minutes for a simple document to a day for a complex report — and it must be redone every time the document is regenerated. That is the argument for doing the work upstream, every time.
Word: use real heading styles rather than manually bolded text, use the built-in list and table features rather than tabs and drawn lines, use a single text flow rather than text boxes. Word's PDF export produces a correct reading order from a well-structured document almost automatically. Text boxes are the notorious exception — they land at the end of the reading order, which is why a "helpful" callout box in a Word document reads at the bottom of the page.
InDesign: build the Articles panel deliberately. Drag frames into it in the intended reading order; that panel is exactly what determines structure order on export. Without it InDesign guesses from frame position and often guesses wrong on complex layouts. Also tick Use Structure for Tab Order in the export options.
PowerPoint: the Selection Pane determines reading order for slide objects, and it runs bottom-to-top — the item at the bottom of the list is read first, which surprises everyone the first time.
LaTeX: the tagpdf package and recent LaTeX kernel work are making genuinely tagged output possible, though it remains more involved than in the GUI tools.
HTML-to-PDF pipelines: the reading order follows the DOM, so well-structured HTML gives you a correct PDF nearly for free. This is one of the strongest arguments for generating documents from HTML — see how to convert HTML to PDF.
Testing without a screen reader
Two proxies that catch most problems in seconds:
Select all and copy. Paste into a plain text editor. What you get approximates the content order — and while it is not identical to structure order, badly interleaved output is a strong signal that something is wrong.
Save As → Accessible Text in Acrobat. This exports following the structure order, which makes it the better test: if the text file reads correctly top to bottom, the reading order is sound. If the page numbers appear mid-paragraph, they have not been artifacted.
Reflow view (View → Zoom → Reflow) reformats the page into a single column following the structure order. A document that reflows into gibberish will read as gibberish.
Why it matters beyond compliance
Reading order is usually framed as an accessibility obligation, and it is one — PDF/UA requires it, and WCAG 2 requires meaningful sequence. But the same structure does other work:
- Text extraction and conversion follow it, so a well-ordered document converts to Word, HTML or Markdown cleanly.
- Search indexing benefits, since extracted text is coherent rather than scrambled.
- Mobile reflow depends on it entirely.
- AI and RAG pipelines ingest documents through text extraction; a document with scrambled order produces scrambled chunks and confidently wrong answers. See building a RAG system with PDFs.
Every one of those is a business reason to get right something that is often treated as a checkbox.
Summary
Structure order — not visual layout, not content stream order — is what assistive technology follows. Check it with Acrobat's Order panel, verify it by exporting Accessible Text or listening to a page, and fix it at the source: heading styles in Word, the Articles panel in InDesign, the Selection Pane in PowerPoint. Get headers and page numbers marked as artifacts, keep captions with their figures, and remember that the same structure powers extraction, conversion, reflow and every downstream system that reads your documents as text.
