AI Citation (LLM Citation)

An AI citation is a link, source card or named mention that an AI search system attaches to a generated answer to show where its information came from. When ChatGPT search, Perplexity, Google’s AI Overviews or AI Mode name your page or brand as a source, that is a citation. MLforSEO treats being cited as the AI-search equivalent of ranking: it is the moment your content moves from being read by the system to being credited in the answer.

How it works

Most AI search systems retrieve pages before they write. A query is expanded into sub-queries (query fan-out), candidate passages are retrieved and filtered, and a large language model synthesises an answer from the passages that survive. Citations are added to the claims in that answer, either inline or in a sources panel. Google’s documentation describes these as supporting links and says a page only needs to be indexed and eligible to show with a snippet to qualify.

Being retrieved and being cited are two separate outcomes. MLforSEO’s material describes a “grounding pool” and a “citation set”: content can be used as grounding context and still never be named. Pages with weak entity recognition, or that only repeat what other sources say, are the ones most often used without credit.

How it’s used in SEO and marketing

• Citation rate as a KPI. Across a fixed set of representative questions, how often is your brand or page named? MLforSEO’s measurement framework uses this in place of position.
• Citation quality. The Agentic SEO course scores each mention from 0 (absent) to 3 (primary recommendation with attributes), because a passing mention in a list of twelve is worth far less than being named as the best fit.
• Gap analysis. Running the same prompts for competitors shows where they are cited and you are not, which usually points to a structural content or entity problem. See co-citation for who you are cited alongside.
• Content and entity work. Extractable passages, specific claims and clear entity signals make it easier for a system to attribute information to you.

Example

A project-management brand runs 15 buyer questions through ChatGPT, Perplexity and Gemini each month. It is named in 2 of 15 answers on one platform (a 13% citation rate), and in both it is listed without attributes, scoring 1. A competitor is described with pricing and team-size fit in most answers. The gap suggests the brand’s pages lack the specific, machine-readable facts the competitor’s pages provide.

Limitations and things to watch

• Citations are not stable. The same prompt can produce different sources from run to run, so track samples over time rather than single checks.
• Citations are not always accurate. The Tow Center at Columbia Journalism Review tested eight AI search tools in February 2025 and found that over 60% of their answers about news excerpts were incorrect, with misattributed sources and broken or fabricated URLs.
• A citation does not guarantee a click, and many citation-driven visits arrive without a referrer (see dark AI traffic).
• Being cited for the wrong use case can bring you the wrong audience, which is why quality scoring matters as much as counting.

Learn it on MLforSEO

• Your rank tracker is lying to you about AI search explains citation rate and citation quality.
• How LLMs Co-Cite: Building Authority by Association covers what earns a named citation.
• LLM Citation Tracking: 4 Step Implementation Guide is a free template for running your own audit.
• The AI Search Optimisation & Agentic SEO course teaches the full citation scoring system.