Table of Contents AI
Open the app: Build bookmarks with AI → (products.documentize.ai)
In one sentence
Table of Contents AI reads a PDF, works out which lines are headings and how they nest, and gives you back the same PDF with a clickable bookmark outline in the reader's navigation panel.
What makes it unique
It is the only app that discovers structure on its own. It first scores every paragraph on font size, boldness, length and spacing to find heading candidates, then asks a language model to confirm them, assign levels and tidy the list, and finally repairs gaps (for example a missing chapter 6 between 5 and 7). Unlike Add TOC you type nothing — but you also cannot edit the result inside the tool.
For long chapters without real headings (a story, for example) the model writes short descriptive titles, and for numbered headings such as "Chapter 6" it keeps the original and adds a dash and a brief summary.
When to use it / when not to
- Use it on a long report, book, manual or manuscript that has no bookmarks yet.
- Use it when headings are inconsistent (different fonts, no numbering) — it uses several signals, not one.
- Do not use it if you need a printed contents page. The site's FAQ is explicit: this tool adds bookmarks, not a visible TOC page. (The site's hero text mentions a contents page too — trust the FAQ and check your result.) For a visible list build one with Add TOC.
- Do not use it on a scan without a text layer: there is nothing to read. Run Searchable first.
Before you start
- PDF only; DOCX and other formats are not supported.
- The PDF should have real text. Password-protected files must be unlocked first (Unlock).
- Page order should be final (see ordering).
Step by step
- Upload your PDF (one or several).
- Leave the two settings as they are for a first run (see below).
- Press the generate button. Progress is shown while the AI reads the document; long books take longer.
- Download the updated PDF and open the bookmark panel in your reader to check the outline.
📷 Screenshot placeholder: the result opened with the bookmark panel visible
Options explained
| Option | What it does | Default | Recommended |
|---|---|---|---|
| Max Layers (1–10) | Intended depth limit for the outline. | 5 | Leave at 5. |
| Only by extracting knowledge | Checkbox shown under the depth field. | On | Leave as is. |
Honest note: in the processing step I read, neither of the two settings is used — the outline depth is decided by the AI from the document itself. Changing them may therefore change nothing (verify on the live app). A third internal behaviour, summarising long section titles, is always on and has no switch on the page.
What you get
- A PDF with the same pages and a nested bookmark outline in the navigation panel; each bookmark jumps to the heading's position on its page.
- For a single document, the outline is also shown on the result screen as a list of titles, levels and pages.
- Nothing visible is changed on the pages: no text, image or layout edits. The original file is untouched.
- With one file the result is a single PDF; with several, a download set.
Quality & limits
- AI-generated structure. Very good on books with clear headings, less predictable on documents where headings look like body text. Always skim the outline.
- Not repeatable to the letter: run it twice and wording or the number of sub-sections may differ slightly.
- If the AI cannot find enough headings or fails, the app falls back to a simple rule that picks the largest text sizes — you still get an outline, a rougher one.
- Free plan: a long document may be processed only up to a point; the app then delivers what it has and offers a continuation to finish later (cooldown 12 hours by default). Each document remembers where it stopped. See If a task stops early.
- You cannot edit the outline here. The site suggests a PDF editor such as Adobe Acrobat.
- Results are kept about 2 days; see Limits & plans.
Works well with
- Use before: Searchable (scans), Merger, Delete Page, Resize.
- Use after: Compressor, Protector.
- Look-alike: Add TOC — manual, with a contents page.
- Recipes: Illustrated ebook, Understand fast.
Common mistakes & fixes
- "No contents page appeared" → by design it creates bookmarks only. Use Add TOC for a page.
- "Headings are missing" → the heading looks like body text (same size, no spacing). Build the entries by hand with Add TOC, or fix the source document.
- "Bookmarks disappeared after I merged/split" → structure-changing apps run first; generate the outline last.
- "A scan gives nothing" → no text layer; make it searchable first.
- "Titles are rewritten" → long text lines are summarised into titles by design.
Same thing elsewhere
- CLI plugin: Documentize.TableOfContents. See CLI.
- MCP tool:
GenerateTableOfContents. See MCP. - API: see API.
Try it
Take a 20-page PDF with chapter titles in larger type. Upload it with default settings, download the result and open the bookmark panel. Count the top-level bookmarks against your real chapters; click three and confirm each lands on the right page.