The percentage of tested AI responses citing a specified business source across a prompt sample.
AI citation rate can describe the percentage of tested AI responses that cite a specified source. For this response-level definition, divide the number of eligible responses containing at least one citation to the target by the total number of eligible responses, then multiply by 100.
A response containing three links to the same target still counts as one citing response under this rule. Counting individual citation occurrences is a different metric and needs a different denominator. Define whether the target is a page, a domain or a documented group of domains before collecting results.
The rate applies to the sample and collection method. Record platform, prompt set, dates and relevant settings. Decide in advance how failed requests, refusals and responses without search features are handled; report exclusions rather than silently removing inconvenient results. If there are no eligible responses, the rate is undefined, not zero.
Takeaway: Use a clearly labeled rate to compare like-for-like samples. Inspect the associated claims and destination pages as well as the number of responses that include them.
Example
A test collects 40 eligible responses and 10 cite the target domain at least once. The response-level citation rate is 25%. If those 10 responses contain 18 links in total, that does not change the response-level numerator to 18.
What to check
Retain raw responses and a target-matching rule. Show numerator and denominator beside the percentage. A higher sampled rate does not establish more customer traffic, accurate citations or increased sales.
For more detail, see Gao et al.: Enabling Large Language Models to Generate Text with Citations.