
Most designers who struggle with a table of contents in InDesign assume the problem is with the TOC feature itself. It is almost never the feature. The TOC panel is a reporting tool: it reads your document and writes down what your paragraph styles tell it to write down. If the styles are inconsistent, the report is unreliable. If the styles are clean, the TOC becomes close to automatic.
That reframing changes how you spend your time. Instead of hand-typing entries and nudging page numbers, you invest in the structural layer of the document — paragraph styles, hierarchy, and naming discipline — and let the software do the mechanical work. Artificial intelligence fits into that structural layer surprisingly well, but not in the way most marketing pitches suggest. AI is not going to magically read a 240-page annual report and produce a flawless contents page. It is very good at the tedious surrounding work: auditing styles, proposing a hierarchy, drafting front matter copy, catching inconsistencies, and writing the checks that stop broken page numbers from shipping.
This guide walks through a complete, practical workflow for building and maintaining tables of contents in InDesign, with AI used as a drafting and QA assistant rather than a replacement for document structure.
Before adding any tool to the process, it helps to understand exactly what happens when you choose Layout > Table of Contents.
InDesign scans the document for paragraphs that use the styles you listed as included styles. Nothing else matters. A heading that was manually bolded instead of styled will never appear. A heading tagged with a local override still appears, but it brings the override with it into the TOC entry.
This is why two identical-looking documents can behave completely differently. One has a locked style sheet; the other is a collection of visual guesses.
Each included style maps to an entry style. The entry style governs indentation, tab stops, leader dots, and whether the page number sits flush right. When designers complain that "the TOC looks wrong," the fix is usually in the entry style, not the TOC dialog.
The TOC reads the page a paragraph actually sits on. If your document uses sections with different numbering — roman numerals for front matter, arabic for the body — the TOC follows those settings. Mismatched numbering is one of the most common causes of a contents page that looks correct until the client reads it closely.
Levels determine tab indentation and sort order. Nested styles and "based on" chains can accidentally pull subheadings into the wrong level. If a TOC suddenly lists every caption in the document, a nested style is usually the culprit.
AI is useful at four specific points in the process, and unhelpful everywhere else.
Export your paragraph style list and paste it into a capable language model. Ask it to group styles by apparent purpose, flag near-duplicates such as "Heading 2" and "Headline 2," and propose a consolidated hierarchy. This takes minutes and typically surfaces three to eight redundant styles in a mature document. Merging those styles is what makes the TOC clean.
When you inherit a document with 60 styles and no logical structure, describe the document type and ask for a proposed hierarchy: part, chapter, section, subsection, sidebar, caption. Then map existing styles onto that hierarchy. The AI suggestion is a starting point, not an answer, but it dramatically shortens the first pass.
Contents pages usually sit alongside other front matter: a preface, an about-this-report note, a how-to-use-this-guide block. Drafting that copy is genuinely faster with an AI assistant, especially when you need a neutral, professional tone across multiple languages.
Ask for a plain-language QA checklist for contents pages: page number alignment, orphaned entries, missing levels, leader dot consistency, cross-reference accuracy. If your team uses scripting, an AI assistant can also draft a script that lists every paragraph using a given style, which makes structural errors visible before export.
Here is the sequence that consistently produces clean contents pages without a pile of manual fixes.
Open the Paragraph Styles panel and read it top to bottom. Delete unused styles. Merge duplicates. Rename anything ambiguous. Decide the exact style names that will feed the TOC and write them down.
A useful rule: if a style name does not tell you where it appears in the hierarchy, rename it. "Heading 2" beats "Blue Bold 14pt" every time.
Use Find/Change with formatting search to locate manually formatted headings. This is the slowest step and the one that determines everything downstream. A single unstyled chapter title means a missing line in the contents.
If the document is long, work chapter by chapter and check the Paragraph Styles panel for locally overridden markers as you go.
In the TOC dialog, include only the styles that belong on a contents page. Map each to a dedicated entry style. Set levels carefully — a common mistake is leaving everything at level 1, which produces a flat list with no indentation.
Build the entry styles as real styles, not local overrides. That way a single edit updates every entry.
Place the contents page as linked text. Never retype it. When the document changes, update the TOC rather than editing entries. Any manual edit inside generated TOC text is fragile and will vanish on the next update.
Compare the TOC against the actual document. Read every entry. Check that page numbers match, that no section is missing, and that numbering styles are consistent. This is the step where AI assistance pays off most: give an assistant the TOC entry list and the document outline and ask it to flag mismatches, ordering issues, and suspicious gaps.
For documents that change weekly or monthly, define an update routine: update TOC, check page count changes, verify that no entry broke, export a fresh PDF, and log the revision. Ten minutes of routine prevents the classic late-night scramble before a deadline.
A contents page rarely lives only in print.
In book panel workflows, generate the TOC at the book level so chapter files report into a single contents page. Attempting per-file TOCs for a multi-file book creates duplicated and conflicting entries.
Reflowable output discards fixed page numbers. A print-style contents page with page numbers is meaningless in EPUB. Build a separate navigation document that uses hyperlinked entries without numbers, and let the reading system handle the rest.
Translated editions break contents pages in predictable ways. German and Finnish section titles expand significantly and wrap onto extra lines. Japanese and Chinese titles are shorter in character count but need different typographic treatment. Arabic and Hebrew require mirrored indentation and right-aligned page numbers.
AI assists with the linguistic side by proposing concise translated section titles, but layout still needs human review. Always check the longest translated title in every language against your tab stops before signing off.
There is no shortage of design assistants. Most are not worth changing your workflow for. Use these criteria.
Short documents with one level of hierarchy rarely need automation. If your contents page has fewer than a dozen entries and the document is stable, generating it manually takes less time than setting up styles. Automation pays off above roughly 20 entries or whenever the document will be revised more than twice.
These are the failures that show up again and again in production files.
A short preflight checklist — styles applied, entries generated, page numbers verified, language variants checked, hyperlinks tested — catches nearly all of these.
Individual discipline breaks down as teams grow, so encode the rules.
Teams that follow this typically cut contents-page production time by more than half, and — more importantly — eliminate the revision loop where a client finds three missing sections after delivery.
| Situation | Recommendation |
|---|---|
| 1–15 entries, stable document | Generate manually; styles still required |
| 20–400 entries, multi-level hierarchy | Automated TOC with locked style sheet |
| Document revised monthly | Automated TOC plus a documented update routine |
| Multi-file book project | Book-level TOC, one master contents page |
| EPUB plus print | Two separate navigation structures |
| Three or more languages | Automated TOC plus layout review per language |
Almost always because the missing headings use paragraph styles that are not included in the TOC style, or because they were formatted manually instead of styled. Open the TOC dialog and confirm the include list, then check the document for unstyled headings.
AI can propose a hierarchy, draft front matter copy, and audit consistency, but the actual page numbers and entry structure must come from InDesign reading real paragraph styles. Anything else will drift the moment the document changes.
In most print documents, yes. A contents page that starts mid-page and continues awkwardly looks unfinished. Reserve a full page, and if entries overflow, use a two-column layout or trim the hierarchy to two levels.
Keep all formatting in entry styles rather than local overrides. Then updating the TOC regenerates entries using those styles and nothing visually changes except the page numbers and any new content.
No. Reflowable ebooks have no fixed pages. Use a hyperlinked navigation document without page numbers and let the reading system present location or percentage instead.
Consolidate first. Export the style list, group and rename the near-duplicates, apply the merged styles with Find/Change, and only then generate the contents page. Generating before consolidation guarantees a messy result.
Read the contents page and the document side by side, following the order of entries rather than the order of pages. Discrepancies stand out immediately because your eye is hunting for the next expected heading instead of scanning randomly.
Run through this list every time a document with a contents page leaves your desk.
A contents page is a promise to the reader: everything listed exists, in that order, at that location. Keeping that promise is a structural discipline, and structure is exactly where a careful AI-assisted workflow pays for itself.

Alexander
Vidu Q4 alternatives compared: Kling 4.0, Seedance 2.5, Wan 3.0, and Runway Gen-4 — 4K support, input modes, durations, audio, and what each costs in credits on Domer.
The top 4K AI image generators on Domer compared: Flux 3 Klein, Nano Banana 2.1, GPT Image 2.5, GPT Image 2, and Nano Banana Pro — 4K credit costs, reference limits, and speed.
Nano Banana 2.1 vs GPT Image 2.5 Flare compared: credit costs per resolution, aspect ratios, reference images, text rendering, and which tasks fit each model.