How to get cited by ChatGPT and AI answers

01How to get cited by ChatGPT: what you can control

How to get cited by ChatGPT is mostly a question of access and clarity. No product publishes a formula for choosing sources, and no markup or file guarantees a citation. What you control is whether the search crawler can reach the page, whether the answer is in the page's text, and whether a passage makes sense when it is lifted out of the page.

A quotable passage is a short piece of text that answers one question completely, with enough context that it stays true when quoted alone. It names its subject instead of saying "it", states the answer in the first sentence, and keeps the supporting detail close by.

This guide is part of the AI answers section. Access is covered in depth in the guide to how ChatGPT and Perplexity read your pages; this page is about structure.

02Start with access

A page that the search crawler cannot fetch cannot be cited. The ChatGPT crawler documentation says sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers, and recommends allowing OAI-SearchBot in robots.txt and allowing requests from the published IP ranges. Blocking GPTBot, the training crawler, is a separate setting and does not remove a site from search answers.

For Google's AI features the equivalent rule, set out in the generative AI optimization guide, is that a page must be indexed and eligible to show with a snippet. For llms.txt, which some services read and Google Search does not, the explainer on robots.txt and llms.txt covers what the file does.

03What does a quotable passage look like?

Compare two openings for an example page about soil testing. The first: "There are many things to consider, and in this article we will explore them." The second: "Test garden soil every two to three years, and before planting a new bed; a laboratory test gives pH and nutrient levels, while home kits give pH only." Only the second can be quoted.

The pattern behind the second opening is simple:

• Answer first. The first sentence under a heading answers the heading.
• Name the subject. Repeat the noun instead of relying on "it" or "this" from an earlier paragraph.
• One idea per paragraph, two to four sentences long.
• Define terms in one sentence that starts "X is", so the definition can stand alone.
• State limits plainly: what is documented, what is not, and the date you checked.

This is not a trick for AI. Google's AI optimization guide says you do not need to write in a special way for generative AI search and that there is no requirement to break content into tiny pieces. It recommends content organized in paragraphs and sections with clear headings, written for people. The same structure serves readers, featured snippets and AI answers.

Images, video and facts in text

The same Google guide recommends supporting text with high-quality images and video where they help. Keep the facts themselves in text, though. A chart can illustrate a trend, but the sentence beside it should state the numbers, and the caption should say what the chart shows and when the data was collected. A fact that exists only inside an image is a fact no text answer can quote.

Dates belong next to any claim that can go out of date. "As of November 2026" costs three words and tells a reader, and anything that quotes the passage, how current the statement is.

04Why do tables and lists get quoted?

Because they hold facts in a fixed shape. A comparison written as prose forces any reader, human or machine, to reconstruct which value belongs to which option. An HTML table states it once, with headers that label every cell.

• Use a real <table> with <th> headers, not an image of a table or a grid of styled divs.
• Give the table a caption that says what it compares and when it was checked.
• Keep one fact per cell and units in the header.
• Use numbered lists for steps in order and bullet lists for items that have no order.

Ranked lists and comparison tables have their own markup, covered in the guide to ItemList schema for ranked lists and tables. Get the HTML right first; markup can describe a table, but it cannot fix one.

05Do headings need to match questions?

They help when they read naturally. A heading phrased as the question a person would type, followed by a direct answer, gives a clear boundary for the passage that answers it. Headings like "Overview" or "Things to know" give no such boundary.

Do not force it. A page made of nothing but question headings reads like a list of search queries. Use question headings where the section genuinely answers a question, and plain descriptive headings elsewhere. Google's guidance on generative AI search also warns against creating separate pages for every variation of a query, which it treats as scaled content abuse.

Keep heading levels in order, too. An H2 for each main question and H3s for the parts of the answer give a page an outline that a screen reader, a search engine and a person skimming on a phone can all follow. Skipped levels and headings used only for styling blur where one passage ends and the next begins.

06Which snippet controls block quoting?

Google's robots meta tag and data-nosnippet specification documents controls that limit what can be shown from a page. They are the opposite of quotability tools, so check that none is applied by accident.

Used on purpose, data-nosnippet is useful for boilerplate you never want quoted, such as a cookie notice or a pricing disclaimer. The same specification notes that the attribute only works on span, div and section elements and that the HTML must be valid, because an unclosed element extends the exclusion to everything after it. These controls are documented for Google; ChatGPT and Perplexity do not document support for them.

07Where do speakable and other markup fit?

Speakable is a schema.org property that points to the parts of a page best suited to being read aloud. Google's speakable documentation describes it as a beta feature for news content, used by Google Assistant to read topical news on smart speakers for English-language users in the U.S. It is not documented as a signal for AI Overviews or for any other answer engine.

The rest of the markup in this handbook plays a supporting role. Article, author and date markup confirms who wrote a passage and when, Organization markup confirms who publishes it, and Product or Review markup states exact values. None of it makes a vague passage quotable. The guide to whether schema markup helps AI Overviews sets out Google's position, and the handbook's overview of structured data and page structure explains how the two fit together.

08A quotable page checklist

1. The search crawlers you care about are allowed in robots.txt and not blocked by a firewall.
2. The main text, headings and tables are in the raw HTML, not loaded by script after the page opens.
3. No nosnippet, restrictive max-snippet or stray data-nosnippet applies to the content you want quoted.
4. Every section opens with a sentence that answers its heading.
5. Key terms have one-sentence definitions.
6. Comparisons are in captioned HTML tables; steps are in numbered lists.
7. Each claim that depends on a source links to that source in the same paragraph.
8. The page shows a byline and an updated date, and the markup repeats both exactly.
9. Nothing important lives only in an image, a tab or a collapsed accordion.