What Actually Changes When a PDF Becomes a Reflowable EPUB3

Converting a pdf to epub is not a rename with a new extension. The fixed page disappears, and what replaces it is a text stream with a sequence the reading system follows, a heading hierarchy it announces, and note links it resolves. How faithful that stream turns out depends less on the converter's interface than on the structure already present in the PDF itself.

This post walks through the anatomy of the conversion in plain terms, so you can predict what happens to your own document instead of watching a progress bar. We start by separating renaming from real conversion, then follow the path from fixed page to reading order, covering tagged PDFs that carry explicit structure and untagged ones where the converter has to infer it. From there: headings moving from font size to a genuine outline, footnote and note structure, tables and figures after conversion, and why the same EPUB3 file can render differently across readers.

The checklist throughout mirrors the reading order, headings and footnote structure that toolkit.bot's preview surfaces, which makes them practical checkpoints rather than claims. The closing sections give you a first ten pages checklist for judging any pdf to epub result.

Renaming a File Is Not Converting

Renaming report.pdf to report.epub changes nothing inside the file.

A PDF stores geometry: text sits at coordinates on a fixed page, and reading order exists only as the sequence a human eye happens to follow. An EPUB 3.3 file is instead a container for HTML, CSS, SVG and other resources, per the EPUB 3.3 Recommendation dated 13 January 2026. No page survives by default.

A real conversion builds that model. It reads the source PDF, decides what each block of text is, and writes a new document with an explicit sequence, heading levels, note links and figure references. That work is invisible in a progress bar.

You therefore cannot judge the result by comparing page 1 to page 1. You read through the sequence and check whether structure survived: does the text proceed as you would read aloud, do headings nest, do note markers link.

A preview shows the difference before you commit. toolkit.bot previews the first 50 pages without an account, enough to judge reading order, headings and notes on most papers and chapters. See why convert PDF files to EPUB.

From Fixed Page to Reading Order

Reading order is not an informal idea. In the EPUB model it is a defined concept with its own section, 4.1 Reading order, in the EPUB 3 Overview, and conversion produces a sequence rather than a page set. That stream reflows: font size, line length, margins and screen width all change how many words fit on a line, so the PDF's page breaks and line breaks mean nothing unless deliberately preserved.

The sequence becomes the file's spine, and the table of contents should match its linear order. The EPUB Accessibility Techniques 1.2 Group Note of 12 September 2026 lists TOC entries matching linear order as a technique, which makes it a check you can run rather than a promise you accept.

This is the biggest difference between the two formats and the first thing to inspect after you convert PDF to EPUB: does the text proceed the way you would read it aloud? For a fuller walkthrough, see PDF Conversion to EPUB: A Practical Guide for E-Readers.

Tagged PDFs: Structure the Converter Can Read

Reading order can be read directly only when the source declares it. A tagged PDF carries explicit structure types: the PDF Association's Tagged PDF Best Practice Guide (v1.0, June 2019) covers block-level elements, illustration elements and table attributes, while PDF/UA-1 requires real content to be tagged in logical reading order.

When those tags exist, headings, lists, tables and figures are declared rather than guessed, and a converter can in principle map them to the corresponding HTML elements with high confidence.

That is why two PDFs of the same report convert very differently: one was exported from a word processor with structure intact, the other flattened by a scan or a print-to-PDF step.

Test it before you convert: in your PDF reader, select a paragraph and copy it. Erratic or empty selection usually means an image-based or untagged file, so structure will have to be inferred. This matters most for journal articles and other long documents, where inferred structure has the furthest to travel.

Untagged PDFs: Structure the Converter Has to Infer

Without tags, the converter works from layout evidence: font size and weight relative to body text, vertical spacing, indentation, page position, and patterns repeated across pages. Every element is a judgment call, and a heading detected as a heading means the converter decided it was one.

Multi-column layouts are the classic hard case. Coordinate data alone carries no reading direction, so column detection depends on whitespace analysis. Why this happens: ambiguous whitespace produces interleaved sentences that read as nonsense.

Scanned documents stack OCR on top of that. toolkit.bot runs automatic OCR on scanned PDFs, but OCR introduces character errors that belong to the source image quality, not the conversion step.

Assess results with that in mind. Inference occasionally makes the wrong call, so verify reading order and notes in the first pages before trusting the rest. Heading levels are subject to the same inference.

Headings: From Font Size to a Real Hierarchy

Inference stops where declared structure begins. In a PDF, a heading is only bigger, bolder text at a coordinate. In an EPUB3 file it should be an h1, h2 or h3 element, and that level drives navigation, screen reader announcements and the generated table of contents.

EPUB Accessibility Techniques 1.2 lists numbered headings reflecting hierarchy as a technique. The frame is useful: if your source numbers its sections, the converted file should reproduce those numbers at the levels the numbering implies.

Watch for these failure modes:

• Running heads and footers captured as headings
• Bold body text promoted to a heading
• A chapter title set as h3 because it was smaller than the previous page's banner
• All headings flattened to one level, so the TOC is a single undifferentiated list

Checking takes roughly two minutes on an e-reader: open the table of contents, then step down one level at a time. Bad nesting shows up immediately in the outline. Why this happens when layout evidence is the only input.

Titles are a separate concern. Accessibility Techniques lists publication and document titles as their own technique, so a converted file should carry a meaningful title. A library view should show the book name, not a filename or an internal identifier.

Footnote and Note Structure: Where the Links Live

A PDF footnote is text at the bottom of a fixed page, positioned by coordinates, with no programmatic link to the superscript in the body. In EPUB3, the note becomes a separate element the body reference points to with a link.

What should change: the superscript marker becomes a link, tapping jumps to the note, and a back link returns you to the exact reading position. With no pages, the note may live at the end of the section or document instead.

Worked example: take a paper with three numbered notes on page 4, note 2 at the bottom of the left column and note 3 at the bottom of the right. After conversion, jump to note 2, confirm the text is correct, then use the back link. Repeat for note 3. If the back link lands in the right paragraph, note structure is sound. If tapping does nothing, the footnote is plain text and you must scroll manually.

Numbers that ran per page in the PDF may need to run per chapter; keeping the original scheme can look broken when read straight through. Table notes and figure captions behave similarly: a caption placed under an image should stay attached to it in the reading order, not float elsewhere.

toolkit.bot states it preserves footnote structure, which this checklist lets you verify rather than trust. Device handling varies too, as stylus notes on an EPUB versus a PDF show.

Tables and Figures After Conversion

Tables and figures raise the same structural question. A PDF table is often just glyphs and rules at coordinates, so nothing states which cell is a header. EPUB3 tables must be expressed as markup rows and cells; the Tagged PDF Best Practice Guide covers table attributes because a tagged table declares header cells and their relationship to data cells. With tagging, conversion can produce a table that reflows, wraps long cell text and stays readable at a large font size. Without it, the converter may fall back to a simpler structure: correct data, cramped layout.

Test the header row first. At your normal reading size, the header should stay tied to its column and cell text should wrap rather than clip. If header and data drift apart, the table was flattened. Wide landscape tables remain a genuine format limitation; rotating the device, viewing the table as a full page, or keeping the original PDF for that table is sometimes the honest answer.

Figures follow the same logic. The Accessibility Techniques list alternative text for images as a technique, so every meaningful figure needs a description while decorative images should be marked as such. OCR reads text, not diagrams, so a scanned chart normally stays an image with, at best, an auto-generated description. Check that figures land in the correct reading order and that informative figures do not carry empty alt text. Review common questions before judging a conversion.

Where the Same EPUB3 File Renders Differently

A table that reads well on one device may clip on another; rendering is the reader's job. Kindle, Kobo, BOOX, PocketBook, reMarkable and Supernote each apply their own typography, hyphenation, margin and embedded-font defaults, and Apple Books or Google Play Books differ again. Treat this as a test matrix, not a ranking.

Convert once, then open the file on the device you actually read on and note three things: does the TOC nest correctly, does tapping a note marker jump and return, and do tables stay readable at your normal font size?

Reader behaviour is not settled: the EPUB Reading Systems draft is a work in progress, so no specification fixes how a note pop-up will appear. Your own test on your own hardware beats it.

Two settings dominate: font size, and publisher versus device layout. If full device layout control visually flattens headings into body text, open the outline view before concluding structure was lost. Structure lives in the markup, and what changes in the EPUB does too.

The First Ten Pages Checklist for Any PDF to EPUB Conversion

Treat this as a practical method, not an industry standard. Nothing published defines a ten page test, but ten pages usually spans the front matter, the first chapter opening, one full text page and the first table or figure, which is where structural problems surface.

1. Reading order. Read the first two pages at your normal font size. If sentences alternate between columns or jump mid-paragraph, stop and inspect the source PDF for a multi-column layout.

2. Headings. Open the table of contents. Confirm the title appears, chapter headings nest under it, and subsection numbering matches the outline levels. Jump to a subsection and confirm the heading still reads as a heading on screen.

3. Notes. Find the first footnote, tap the marker, read the note, then use the back link. Confirm the note is numbered the way it was in the source.

4. Tables and figures. Jump to the first table. Check the header row still reads as a header and cells wrap rather than clip. Check the first figure keeps its caption in place and, if it carries information, some description text.

5. Metadata. Confirm the title and author in your library list are the document's real title and author, not the filename.

Five passing checks mean the rest is likely sound, because these exercise the conversion decisions that repeat. If you are deciding whether to convert at all, run the preview first: toolkit.bot previews the first 50 pages without an account, and a free account covers full documents, so you can run the test before committing to a whole book.

Keep the original PDF. A converted file is a second representation of the work, and long tables, complex equations and scanned charts sometimes remain easier to read in the fixed layout you started with.

What to Do With What You Now Know

You can now predict a conversion instead of watching a progress bar. The page grid disappears, a reading order takes its place, and headings, notes, tables and figures become structural objects the converter must build or infer.

Act on three things: check whether your PDF is tagged before converting, since tagged sources carry structure a converter can map rather than guess; run the five point check on the first ten pages instead of judging the file after an hour; and test on your own device, because Kindle, Kobo, BOOX, PocketBook, reMarkable and Supernote render the same file differently.

When a conversion passes, keep both files. The EPUB3 is your reading copy; the PDF stays your reference copy for anything the reflow cannot fully express.

Conclusion

A PDF to EPUB3 conversion is not a rename. It replaces a fixed page grid with a reading order, and it turns headings, notes, tables and figures into structural objects the converter maps or infers. Tagged PDFs give that structure directly; untagged files require guessing. Your best control is process: check tagging before you convert, run the five point check on the first ten pages, and test the result on your own e-reader, because rendering varies across devices. Your next step is simple: preview the first 50 pages with toolkit.bot before committing, then keep both the EPUB3 reading copy and the original PDF reference. You can now predict the outcome instead of hoping. Run that preview on your next document and let the structure, not the progress bar, guide your decision.