What is an AI citation?

Citation versus mention

You can be mentioned without being cited, when the answer names you from other sites' content, and cited without being mentioned, when your article supports a general fact. A recommendation query such as "best espresso grinder for beginners" mostly produces mentions. A factual query such as "how fine to grind for espresso" mostly produces citations.

Where citations appear

• Google AI Overviews and AI Mode: link cards and inline links next to the generated text; see AI Overviews.
• Assistants with web search, such as ChatGPT, Claude, Copilot, Gemini and Perplexity: numbered references or source chips, typically with a list of sources at the end.
• Voice assistants: a spoken attribution, "according to…", with no link.

Formats change often. The constant is that citations come from pages the system retrieved at answer time, a process described under retrieval-augmented generation.

What makes a page citable

No engine publishes a formula. What follows is how the retrieval process works, plus Google's own statement that its AI features need no special optimisation beyond what it asks for in Search.

1. Access. The engine's search crawler can fetch the page, and the page is indexed. Check robots.txt, firewalls and noindex.
2. Relevance to the specific question. Assistants break a prompt into narrower searches. A page that answers one of those precisely is more likely to be retrieved than a broad overview.
3. Statements that stand alone. A sentence containing a complete, specific fact is easy to attribute.
4. Something not available everywhere. Original data, first-hand testing and precise definitions give a system a reason to cite you instead of any of ten similar pages.
5. Clear authorship and upkeep. Dates, named organisations and accurate, current facts support trust in the source.

In Google, a page must also be eligible for a snippet. nosnippet keeps it out of AI Overviews and AI Mode.

Measuring citations

This is harder than rank tracking, for three reasons. Answers vary between runs of the same prompt. They are personalised by location and history. And there is no public "position".

What you can observe:

• Referral visits. In your analytics, look for referrers such as chatgpt.com, perplexity.ai, copilot.microsoft.com and gemini.google.com. This undercounts, since many people read the answer and never click, and some apps pass no referrer.
• Search Console. Google includes clicks and impressions from AI Overviews and AI Mode in the Performance report totals under the Web search type. It does not currently break them out as a separate filter.
• Prompt sampling. Run a fixed set of customer questions against each assistant repeatedly and record how often you are cited and mentioned. Because single runs are noisy, the figure that means something is a rate over many runs.

Hydrogen SEO's AI visibility feature, on its Pro and Max plans, automates the sampling: you supply questions and competitors, and it reports visibility, citation rate and share of voice with a confidence range; see answer engine optimization. The method is explained in how to measure AI visibility.

Citations and traffic

A citation is visibility first and traffic second. Many readers take the answer and move on, which is the same pattern as zero-click searches. Those who do click have usually read a summary already and arrive with a specific intent. Judge the channel by the quality of those visits and by whether your brand is present in answers about your field, not by raw session counts. Steps for earning more are in how to get cited by AI.