# Sectioning (/docs/convert-pdf/sectioning)



A book is more than a sequence of pages — it has chapters, headings, paragraphs, activities, figures. Sectioning is where that structure is identified and confirmed.

Sectioning is its own stage, run after [extraction](/docs/convert-pdf/extract) — the AI labels chapters, headings, and pedagogic content here, not during extraction itself. Once it's run, this is your moment to review that work: check that the structure matches the real book, fix what was misread, and clean out what doesn't belong.

## Why it matters [#why-it-matters]

The structure defined here drives almost everything that comes later:

* The **table of contents** is built from it
* **Navigation** for readers (including screen reader users) depends on it
* **Quizzes and glossary** are generated per section, so wrong boundaries produce mismatched content
* **Narration** follows the reading order it defines

A few minutes of review here saves much more correction time later.

## The sectioning screen [#the-sectioning-screen]

For each page, you get a split view: the original page image on one side, and an editable tree of that page's content on the other. Every piece of content — a heading, a paragraph, an activity, an image — is a node in that tree.

* **Structure** comes from where a node sits in the tree, not from a numbered heading level. There's no H1/H2/H3 — a node is nested under whichever node it belongs to, and marked with a role (like "heading") through a dropdown on the node itself.
* **Reading order** is just the order nodes appear in the tree. Drag a node by its handle and drop it where it belongs to reorder content.
* **Noise removal** is a per-node toggle — an eye icon that excludes a node from the rendered output without deleting it. Repeated headers, footers, and page numbers are excluded by default, but check that nothing else slipped through (or got excluded by mistake).
* **Activities and special content** get their own types — a per-section dropdown lets you mark something as, say, a matching activity or a multiple-choice question, rather than leaving it as plain text.

A row menu on each node also lets you add text, add an image, group nodes together, duplicate, or delete — for the cases where the AI missed something or split content incorrectly.

<Callout title="Bird's-eye view">
  Switch to **Overview** to see every page's sections in one table — page, section, type, content, and parts — instead of reviewing one page at a time.
</Callout>

## What to look for [#what-to-look-for]

* **Is the structure right?** A node nested in the wrong place, or given the wrong role, changes the hierarchy of everything beneath it.
* **Is the reading order correct?** Especially in two-column layouts, check that content flows in the order a reader should encounter it.
* **Is there leftover noise?** Repeated headers, footers, and page numbers are usually excluded automatically — but double-check, and exclude anything else that shouldn't be read aloud or shown.
* **Are activities and special content recognised?** Exercises, boxes, and figures should carry the right type, not sit as plain text.
* **Does a page carry a "no text layer" warning?** A flag on a page means its PDF text layer was empty and the content shown was recovered from the page image instead — give those pages a closer read, since vision-based recovery is more error-prone than reading embedded text.

## When you're done [#when-youre-done]

With the structure confirmed, the visual side of your ADT comes next: [Storyboard](/docs/convert-pdf/storyboard).