What headings are recognized?

It recognizes ATX-style markdown headings — lines beginning with one to six hash characters followed by a space, such as

What are the word counts for?

Each section's word count is the number of words between its heading and the next heading in the document. This shows you where the bulk of a long response lives, which is handy for spotting padding or finding the substantive parts quickly.

Can I copy the table of contents?

Yes. You can copy the whole outline as an indented markdown bullet list to paste into a document, with each heading nested by its level.

Does it work if there are no headings?

If the response has no markdown headings the tool tells you so rather than inventing structure. In that case the response is a single flat block; consider prompting the model to use headings, or use a key-point extractor instead.

What is the LLM Output Section Navigator?

Parses markdown headings from LLM output and builds a nested table of contents with heading levels and word counts per section. Makes long AI-generated documents easy to scan and navigate. It runs free in your browser on Gera Tools, with nothing uploaded.

LLM Output Section Navigator

Name: LLM Output Section Navigator
Creator: Gera Tools
License: https://creativecommons.org/licenses/by/4.0/

Get one useful tool a week

Like this tool? Enter your email and we'll send you one genuinely useful Gera tool a week — plus a link to come back to this one. No spam, one-click unsubscribe any time.

Turn a long AI document into a navigable outline

LLMs love to produce long, multi-section documents — reports, guides, plans — and scrolling through them to find the part you want is tedious. If the response uses markdown headings, this tool parses them into a nested table of contents with a word count per section, so you can see the document’s shape at a glance and jump straight to what matters.

The per-section word count is more informative than it first appears. It acts as an automatic quality signal: a well-structured document distributes words roughly proportional to the importance of each section. When the word counts are wildly uneven, that asymmetry is usually telling you something about the model’s output quality.

Patterns worth watching:

Thin implementation sections. A 20-word “How to apply this” following a 500-word “Background and Theory” is a common LLM failure mode: fluent context, vague action. The word count ratio makes this visible at a glance.
Bloated introductions. The model restates the question at length before getting to substance. The first section will have a disproportionately high count.
Redundant conclusion. If the “Summary” section is nearly as long as the main content sections combined, the model has repeated everything rather than condensed it.
Missing sections. A section heading with zero or near-zero word count means the model created the heading but left the content empty — a structural placeholder without substance.

How it works

The navigator scans each line for markdown heading syntax. It detects ATX headings, where one to six leading hash characters set the level, and Setext headings, where a line is underlined with equals signs (level one) or hyphens (level two). Each heading becomes an entry indented by its level to form a nested outline. For every section it counts the words between that heading and the next heading in the document, giving you a sense of where the content is concentrated. The result is a clean, copyable table of contents.

Practical uses

Before sharing a generated report: check the outline to confirm all expected sections exist and are substantive.
When editing AI drafts: navigate directly to thin sections that need human expansion.
When comparing two responses to the same prompt: the outline shows structural differences between them at a glance.
Before inserting a TOC into a document: copy the outline and paste it at the top in one step.

Tips and notes

Use the per-section word counts to spot where a response is padded versus where the real content lives — a guide with a 400-word “Conclusion” and a 30-word “Implementation” section is telling you something. You can copy the full outline as a markdown list to drop a TOC into your own document. If a response has no headings the tool says so rather than guessing; if you want structure, prompt the model to use markdown headings or run the output through a key-point extractor instead. Everything runs locally and updates as you type.