Answer in brief
Citation rate is the share of a declared set of eligible AI answers that visibly cites a specified source, domain, or URL. Report the numerator, denominator, prompt cohort, products, and observation window—or the percentage cannot be interpreted responsibly.
Start with one explicit counting rule
For a target-source citation rate, the numerator is the number of eligible answers that visibly cite the declared source target. The denominator is every eligible answer collected under the same protocol. Multiply that ratio by 100 to express it as a percentage.
The target must be written down before collection begins. A brand mention, a link to the brand’s domain, a link to one exact page, and a third-party article about the brand should not be folded into the same event.
Freeze the cohort before comparing results
Build a fixed panel of prompts that represents the questions you care about, then declare the products, visible model or surface, locale, date range, and eligibility rules. Preserve the prompt, answer text, displayed URLs, and timestamp for every observation.
Run the same panel repeatedly. A movement in the rate is meaningful only when the underlying cohort remains comparable and known product or methodology changes are annotated.
Keep mentions, citations, and links separate
A brand can be named without its website being cited. One answer can also show several links from the same domain. Store a yes-or-no answer-level citation field for the rate, then retain each displayed URL in a separate evidence record.
Review whether a cited page actually supports the nearby claim. A visible link establishes attribution, not endorsement, accuracy, or recommendation quality.
Report by product before blending
Source behavior can differ across answer products. Show each product’s result first using its own eligible-response count. A blended rate can summarize the declared panel, but it should never conceal the platform-level table.
Pair citation rate with mention rate, recommendation context, source quality, and competitor presence. The result is a measurement of observed answers under stated conditions—not a universal authority score or a view into a proprietary ranking formula.
Data behind the finding
| Measure | Result | Context |
|---|---|---|
| Target citation rate | Cited answers ÷ eligible answers | Counts answers that visibly cite the declared source target |
| Mention rate | Mentioning answers ÷ eligible answers | A name can appear without a link to an owned source |
| Raw link count | All displayed links | Useful supporting evidence, but not interchangeable with cited-answer rate |
AthenaHQ generated the initial brief from three tracked prompts about citation-rate definitions, improvement, and monitoring. The published version was rewritten around a declared response-level counting rule, checked against primary OpenAI and Google documentation, and stripped of vendor-comparison claims that were not necessary to answer the question.
How to cite this article
A canonical source for this finding
Plum, Jenna. “What is citation rate in AI search—and how should you measure it?” The Answer Signal, August 13, 2026.
https://theanswersignal.com/blog/citation-rate-ai-searchSources and related research
- Introducing ChatGPT searchOpenAI. Primary product documentation describing source-linked answers and the Sources panel
- Publishers and Developers FAQOpenAI. Primary guidance on discovery, citation links, crawler access, and referral measurement
- Optimizing your website for generative AI features on Google SearchGoogle Search Central. Primary guidance on how generative Search features relate to core Search systems