How to get cited by ChatGPT and Perplexity

To give an AI assistant a page it can cite, put a direct answer on the page that owns the topic and keep that page public and crawlable. Then check the cited links in live answers. A citation is a source link, not proof that the assistant recommended the brand.

That distinction matters. Our guide to mentions and citations explains which surfaces produce each one and what the default LLM Audit measurement covers.

Put the answer on the page that owns it

Our own sample covered 505 live buyer answers and 438 cited URLs over 60 days. Of those URLs, 400 pointed to a home page and 38 to product, pricing or service pages. None pointed to a path our classifier recognized as a FAQ, help center, blog or news article.

The sample has limits. It classified pages by URL pattern, and only ChatGPT and Gemini returned URLs in those answers. It does not prove that assistants never learn from a FAQ. It does show where they sent the buyer in this sample: the home and product pages.

The method and caveats are in FAQ page or H2 sections: what AI models actually cite.

Use that result as a placement rule:

• Put the basic description and audience on the home or product page.
• Put price, inclusions and exclusions on the public pricing page.
• Put an integration answer on the integration or product page.
• Put a comparison on a page that names both options and says who each one fits.

Make each answer easy to extract

Use an H2 that matches the buyer's question, then answer it directly in the first paragraph below the heading. Keep the answer self-contained: name the product, state the relevant fact and include the condition that changes the answer.

A short answer is useful only when it is complete. Do not remove a price, location, limit or exception to hit an arbitrary word count. The page still has to work for the person making the decision.

The questions to cover should come from the buying process, not a keyword list. The buyer-intent prompts page defines the category, comparison, objection, use-case and context questions that belong in a measurement.

Support claims with sources

Link a factual claim to its original source. Name the study, documentation or dataset instead of writing "research shows". For example, the Princeton GEO study by Aggarwal and colleagues reported visibility gains of up to 40 percent from methods that added citations, quotations and statistics in its benchmark.

The same paper found little to no improvement from keyword stuffing. That is a reason to make a claim checkable, not a reason to add statistics that do not belong on the page.

Keep the page reachable

The answer has to be available without a login wall. Serve the meaningful text in the page response, use a canonical URL and avoid blocking the crawlers you want to reach the content. If a bot cannot fetch the page, that page cannot be the linked source in its answer.

Show an updated date only when the content was actually reviewed or changed. A date is useful evidence of maintenance when it reflects real work.

Measure the cited answer

Run the same buyer questions after publishing. Keep two counts separate:

1. Mentions: answers that name the brand in the text.
2. Citations: links the answer uses as sources.

A cited page can support an answer that recommends another brand, and a model API without web search can name a brand without linking to any page. Reporting the two as one metric hides that difference.

The practical workflow is simple: answer the buyer's question on the page that owns it, make the facts checkable, keep the page reachable and inspect the live answer rather than assuming the change worked.

---

Want to see which open buyer questions name your brand or a competitor? Run the free audit at llmaudit.app. It asks the models live and reports the answers as counts with their total.