Live on Product Hunt Intraface launched today. Less steering, more vacations. See the launch
Start for free

What are AI citations?

AI citations are the links an answer engine attaches to the sentences it writes, pointing at the pages it drew on. ChatGPT, Perplexity and Google's AI Overviews all show them, each in its own way. Being cited is the AI-search equivalent of ranking: your page is named inside the answer instead of listed under it.

Part of Glossary: the terms behind MCP, coding agents and shared pages, defined plainly.

How it works

You ask, the engine searches and reads a handful of pages, then answers with numbered links back to the ones it used.writes the answerindexed, readable textyouanswer engineChatGPT, Perplexityweb searchyour page1. question2. searches, fans out3. fetches the top hits4. answer with [1] [2]
  1. You ask a question. The engine decides it needs fresh facts and runs one or more web searches, often several across subtopics.

  2. It fetches the top results and reads them, keeping what answers the question. Only a slice of each page gets used.

  3. It writes the answer and marks which span came from which page. OpenAI's API returns url_citation annotations with url, title, start_index and end_index.

  4. Perplexity shows numbered markers like [1] and returns search_results with title, url, date and snippet. Google's AI Overviews show links beside the text.

  5. Inline citations are a subset. OpenAI's sources list is every URL the model consulted; only the most relevant make it into the answer.

Source: OpenAI: web search tool, citations

AI citations versus AI visibility

People use the two terms together, and keyword tools lump them. They are not the same. AI visibility is how often an engine mentions you at all: your name in a comparison, your product in a list, your page as a link. A citation is the narrow, checkable case: a link in the answer that points at one URL of yours. You can count citations. You can only estimate visibility, usually by asking the engines a fixed set of questions and recording what comes back.

Ranking is the other neighbour. In classic search your page is listed and the reader chooses. In an AI answer the engine chooses for them, uses a slice of your page, and may or may not put the link where anyone sees it. Being cited still beats being absent, because the reader who clicks has already read your point.

Example: one question, three engines

Ask "what is a remote MCP server" in three places and the citation plumbing differs.

ChatGPT, through OpenAI's Responses API, returns the answer text with annotations. Each is a url_citation carrying a url, a title, and a start_index and end_index that mark the span of text the page supports. OpenAI requires apps that show web results to make those citations clearly visible and clickable. A second list, sources, holds every URL the model consulted, and it is usually longer than the citations.

Perplexity answers with numbered markers in the text, [1], [2], and returns a search_results list with a title, URL, date and snippet for each source. The markers in the answer point at entries in that list.

Google's AI Overviews show links alongside the generated text. Google says there are no additional requirements to appear in them and no special markup, but it does not publish how a page gets chosen for a given overview.

Why it matters

Engines read a few pages per question and quote a sentence or two from each. That changes what a page has to do. The answer has to be stated, not implied, and stated early, because the model works from retrieved text with a budget. OpenAI's search_context_size controls how much of the search results the model sees, and the docs say it does not guarantee any number of sources or citations. A page that takes four paragraphs to reach the definition gives the model nothing quotable in the part it reads.

The glossary pages on this site open with a 40 to 60 word answer for exactly this reason: the definition is the first thing a crawler or a model sees, and it stands on its own.

Common mistakes

  • Counting mentions as citations. A brand name in an answer with no link is visibility, not a citation. Track them separately or the numbers mislead.
  • Blocking the crawler, then wondering. If robots.txt blocks an engine's crawler, that engine cannot read the page, and an unread page is never cited. Check the rules against each engine's documented user agent.
  • Writing for the summary, not the click. A page that is quoted but never visited still earned its place in the answer. Make sure the page the link lands on gives the reader a reason to stay.
  • Assuming the published mechanics are complete. OpenAI documents the format of a citation, not the ranking behind it. Google documents the controls, not the selection. Treat any vendor's "how to get cited" list as a hypothesis.

FAQ

Are AI citations the same thing as AI visibility?
No. AI visibility is the broader idea: how often your brand or site shows up in AI answers at all, as a link, a name or a quote. A citation is the concrete, countable form of it: a link in the answer pointing at one page of yours.
How does ChatGPT choose which pages to cite?
OpenAI does not publish the ranking. Its docs say the web search tool returns results based on top results, the model reads them, and inline citations show only the most relevant references, while the sources list holds every URL it consulted. A page that states the answer in plain text near the top gives the model something it can quote.
What is the difference between ChatGPT citations and sources?
Citations are the links shown inline in the answer. Sources is the full list of URLs the model retrieved while searching, which the API returns when you ask for web_search_call.action.sources. The sources list is usually longer, because consulting a page does not mean citing it.

Stop sending HTMLover Slack.

Create an account in a minute, connect your coding agent, and make your first board.