Citation share

The metric needs three defined inputs: a prompt panel, an engine list and a counting rule. The panel is a fixed set of real buyer questions in your category, kept stable so results are comparable over time. Engines are tracked separately, because citation behavior differs sharply between ChatGPT, Perplexity, Gemini and Google's AI surfaces. The counting rule states what a citation is: most implementations count each cited link in each answer, then compute your domain's percentage of the total, sometimes alongside a simpler figure, the percentage of answers citing you at least once.

Repetition is part of the method rather than a refinement. Generative answers vary between identical runs, and SparkToro's research found under a 1 percent chance that two identical ChatGPT runs name the same brands. A citation share computed from one pass over the panel is mostly noise; computing it across many runs per prompt, on a schedule, yields a stable figure whose movement means something.

Citation share is one of three distinct visibility metrics, and conflating them causes bad decisions. Mention rate counts how often your brand is named in answer text, cited or otherwise. Share of voice compares your mentions against competitors' mentions. Citation share tracks links, meaning your domain served as evidence. The metrics diverge routinely: engines often name brands from memory without citing them, and cite pages, especially third-party reviews and comparisons, without naming the host brand in the answer.

The baseline distribution is heavily skewed toward a few aggregator domains. 5W Research's analysis of ChatGPT citation share found Wikipedia and Reddit together exceed 25 percent of citations, with review platforms and comparison sites also over-represented. For most brands, the practical ceiling on own-domain citation share is modest, which is why serious programs track citations of favorable third-party pages alongside their own domain.

Citation behavior also differs by engine, which is why per-engine measurement matters. Otterly.AI's Citations Report, built from over a million data points, documents distinct source preferences across ChatGPT, Perplexity and Google's AI surfaces, and cross-platform analyses find only partial overlap in the domains different engines cite for the same questions. A brand can hold a healthy citation share in one engine while sitting at zero in another, and a blended number would hide exactly the gap that needs work.

Citation patterns also diverge from organic rankings. Ahrefs found 28.3 percent of ChatGPT's most-cited pages had zero Google organic visibility, and ALM Corp measured the share of Google AI Overview citations drawn from top-10 organic results falling from 76 percent in July 2025 to 38 percent by March 2026. Engines increasingly cite whatever best answers the specific sub-question, so rank tracking predicts citation share poorly and cannot substitute for measuring it.

The absence problem is widespread enough to be the headline finding in category-level studies: one 2026 analysis found 51 percent of B2B tech brands had zero citations across ChatGPT, Perplexity and Gemini. For those brands, citation share is a discovery metric before it is an optimization metric, since the first measurement establishes whether the brand exists in AI answers at all.

Growth starts with retrievability: pages that AI crawlers can fetch, indexed in the corpora each engine grounds against, with one clear question answered per page. From there, the strongest documented lever is making passages function as evidence. The Princeton GEO study at KDD 2024 found that adding statistics, quotations and cited sources lifted generative engine visibility by up to 40 percent, because engines cite pages that give them something concrete to attribute.

The second lever is presence on the domains engines already prefer. Since aggregators, review platforms and comparison pages absorb a large share of citations, a complete and current profile on the third-party sites in your category often adds more cited surface area than another page on your own domain. The goal shifts from citations of you to citations that carry you: every cited page that presents your brand accurately counts commercially, wherever it lives.

Operationally, citation share belongs on a monthly dashboard next to mention rate and share of voice, per engine and per prompt cluster, with the raw answers stored so every data point can be audited. Reachroller tracks this against ChatGPT live through the official API with web search enabled and scores from stored evidence, with other engines rolling out. Whatever tooling you choose, distrust any citation share figure that cannot show you the underlying answers.