Write content that gets cited

Documentation Index

Fetch the complete documentation index at: https://docs.getcited.in/llms.txt

Use this file to discover all available pages before exploring further.

Not all content is equally citable. AI platforms prefer content that is structured, definitive, and self-contained — content where a single paragraph can answer a question without needing surrounding context. The single highest-leverage pattern is the definition-first lede: starting every page with a 1-2 sentence definition of the topic. This playbook covers that pattern and six others that increase citation likelihood, based on observed AI platform behavior and citability testing across the Cited Index dataset.

​

The citability principle

LLMs extract and cite content at the paragraph and sentence level, not at the page level. A page can have excellent content overall, but if no single paragraph stands alone as a complete answer, the LLM has nothing quotable to cite. The goal is to write so that any paragraph on your page could be extracted and used as an AI answer — each one readable without the paragraphs around it. This is the same principle that makes Wikipedia the most-cited source across AI platforms: every Wikipedia paragraph is written to be self-contained. You do not need to read the preceding section to understand the current one.

​

Seven patterns that increase citation

These are the practitioner version of the content patterns observed across highly-cited pages in the source preference hierarchy. 1. Definition-first ledes. Start every page with a 1-2 sentence definition of the topic. “Mention rate is the percentage of tracked prompts for which an AI platform mentions a brand by name.” This is the single most-cited sentence pattern — LLMs disproportionately pull from the first paragraph of a page. 2. Entity before data. Put the brand or concept name before the number. “The median Indian brand has an 8% mention rate” is more citable than “8% is the median mention rate for Indian brands.” LLMs associate data with the entity that precedes it, so entity-first phrasing produces cleaner citations. 3. Pre-summarize tables. Before any data table, write one sentence that captures the main finding. LLMs often cannot parse tables directly — the pre-summary gives them a citable sentence. For example: “Consumer categories outperform B2B categories at the median, led by Travel and Luggage at 19%.” 4. No hedging on core claims. “Mention rate is the most fundamental AI visibility metric” is citable. “Mention rate might generally be considered one of the important AI visibility metrics” is not. Be definitive where the claim is defensible. Hedged language signals uncertainty, and LLMs rarely cite uncertain statements. 5. Self-contained FAQ answers. Each FAQ answer should read standalone without needing the question. LLMs can extract answers independently of the question-answer pair. An answer that says “It depends — see the section above” is uncitable. An answer that restates the context and provides a complete response is citable. 6. Explicit comparisons. Put both values in the same sentence. “ChatGPT shows a median mention rate of 10%, while Perplexity’s median is 5%.” Not: “ChatGPT shows higher rates than Perplexity.” Vague comparisons force the LLM to look elsewhere for the actual numbers. 7. Key findings at the top. The first paragraph should contain the single most important claim. LLMs disproportionately cite from the first paragraph of a page — burying the key finding in section four means it is less likely to be extracted and quoted.

​

What NOT to do

Five content patterns actively reduce citation likelihood:

• Long introductions before the actual content starts. LLMs skip preamble. If your first three paragraphs are scene-setting before you make a substantive claim, the LLM may never reach the claim.
• Marketing language. Phrases like “best-in-class,” “revolutionary,” and “industry-leading” signal promotional content. LLMs avoid citing promotional language because it is not independently verifiable.
• JavaScript-rendered content that does not appear in raw HTML. AI crawlers parse HTML, not rendered JavaScript. Content loaded via client-side React, Angular, or Vue is often invisible to crawlers. Server-side rendering or static generation is required for crawlability.
• Paywalled or login-required content. AI crawlers cannot authenticate. Content behind login walls or hard paywalls is invisible to citation.
• Thin pages with fewer than 300 words. Pages without enough substance to be authoritative are rarely cited. A 150-word page that says “we offer CRM software” is not citable; an 800-word page that explains how CRM categories differ is.

​

How to audit your existing content

A three-step audit that takes 30-60 minutes for your five highest-priority pages:

1. First-paragraph test. For each page, read the first paragraph and ask: “Could an AI quote this paragraph as a complete answer to a question?” If the answer is no — because it is vague, hedged, or requires context from elsewhere on the page — rewrite it with a definition-first lede.
2. Table pre-summary test. For each table on the page, check: is there a natural-language sentence before it that summarizes the main finding? If no, add one. This single change can make previously-uncitable data suddenly quotable.
3. Self-containment test. Pick any paragraph at random from the page. Read it in isolation. Does it make sense without the paragraphs before and after it? If it starts with “This” or “It” referring to something in a prior paragraph, rewrite the opening to name the entity directly.

This audit typically produces immediate improvements. The changes are small — a rewritten first paragraph, a pre-summary sentence, a pronoun replaced with an entity name — but they make the difference between content that AI platforms can cite and content they skip.

​

Expected impact

Content structure improvements typically show mention rate effects within 4-12 weeks as AI platforms re-index and re-process content. Retrieval-first platforms like Perplexity may reflect changes faster, within 1-2 weeks, because they fetch content in real time rather than relying on periodic training data updates. Training-data-dependent platforms like Claude and base ChatGPT take longer because content changes propagate only when the model’s training data is refreshed.

​

Related concepts

• What sources LLMs cite
• Citations vs mentions
• Mention rate
• Schema markup

​

Frequently asked questions