--- language: en title: "Citation recall" description: "Citation recall measures the proportion of verifiable claims in a response that are fully supported by the cited sources. Learn how to define your baseline." canonical: "https://tropk.ai/en/resources/glossary/citation-recall" last_modified: "2026-10-12T16:13:01.776Z" --- # Citation recall Reviewed on 2026-10-06. ## What it means [Citation](/en/resources/glossary/ai-citation) recall is the proportion of verifiable claims in a response that are fully supported by the associated references. In the method by Liu, Zhang, and Liang, the unit analyzed is a sentence about the external world. The denominator includes all sentences that require verification, including those that appear without any citations in the response. ## Why it matters for marketing This measure makes it possible to assess evidence coverage when an AI describes products or compares suppliers. A response can include several links and still leave commercial terms unsupported. For marketing, distinguishing supported claims from the rest helps identify which information needs investigation before using the response as a basis for content or customer service. ## Example in practice Consider a hypothetical response about an event platform with ten verifiable sentences. Six are fully supported by the associated sources; two have partial support, and two have no references. Under this protocol, citation recall is 60%. The team decides to review the remaining four sentences before reusing the text in a sales presentation. ## How to apply or verify Divide the response into sentences and identify which ones require verification. Open the cited documents and assess whether they support the entire sentence, considering the references together when necessary. Record the numerator, denominator, and support criterion. An inaccessible source requires an evaluation note; lack of access should not automatically become a factual conclusion. ## Important distinctions Citation recall differs from citation precision, whose denominator is the references, and from a brand’s [citation rate](/en/resources/glossary/citation-rate). The 51.5% reported in the study is the average coverage across four engines examined using responses collected in 2023. This figure does not mean that 51.5% of all [AI responses](/en/resources/glossary/ai-response) are false. ## Sources and references - [Evaluating Verifiability in Generative Search Engines]() — Liu, Zhang e Liang, arXiv ## Continue a leitura - [AI citation]() - [Citation faithfulness]() - [Citation rate]() ## About the author [Vinicius Pacheco]() Responsible for Tropk’s technical content, focusing on crawling, indexing, structured data, and agent access to content. --- [All pages in Markdown]()