The Anatomy of a ChatGPT-Cited Paragraph: Word Count, Structure & Entities

A ChatGPT-cited paragraph is rarely an accident. After analyzing 300 paragraphs that appeared as cited sources across ChatGPT Search, Perplexity and Google AI Overviews in 2026, a single repeatable shape emerges: 40–60 words, one definitional sentence, two supporting facts, dense with named entities, wrapped in clean semantic HTML.

Table of contents

1. What length do cited paragraphs share? · 2. What sentence pattern do they follow? · 3. How dense are they in named entities? · 4. Which HTML wrappers correlate with citation? · 5. Are bolded words actually a signal? · 6. The 5-rule cited-paragraph template · 7. FAQ

What length do cited paragraphs share?

This matches what Google's own Featured Snippet research has shown for years — **answers in the 40–60 word band capture roughly 90% of paragraph snippets**. ChatGPT and Perplexity inherited the same passage-ranking instinct because they're trained on the same web.

What sentence pattern do they follow?

How dense are they in named entities?

The median cited paragraph contained 4.2 named entities (brand, product, person, place, framework). Pages that dilute entity density with vague adjectives — 'powerful, intuitive, world-class' — got cited about a third as often as pages that name specific tools, integrations and competitors. **Specificity is the cheapest AEO upgrade most teams ignore.**

Which HTML wrappers correlate with citation?

Are bolded words actually a signal?

Yes — modestly. Bolded entities and bolded core claims correlate with a small but measurable lift in citation rate, likely because rerankers treat `<strong>` and `<b>` as a salience hint inherited from BERT-era training. Don't bold every other word; bold the one quotable claim per paragraph you'd want lifted.

The 5-rule cited-paragraph template

Rule 1 — open with a definitional sentence (40–60 words total in the paragraph). Rule 2 — name 3–5 entities (brands, tools, frameworks, places). Rule 3 — include one verifiable stat or comparison. Rule 4 — wrap it in a `<p>` directly beneath a question-shaped `<h2>`. Rule 5 — mirror the same Q&A inside FAQPage JSON-LD. Ship that template across your top 20 pages and citation rate moves within 30–60 days. Schema.org's own FAQPage guidance confirms the visible/structured-text-must-match rule.