Definition
Citation rate is the percentage of AI responses in a tracked prompt set that cite your domain at least once as a source. If ChatGPT produces 200 responses to your prompts in a week and 18 of them link to your site, your ChatGPT citation rate for that week is 9%. It answers a simple question: how often does this engine reach for my content at all?
The formula is responses citing your domain, divided by total responses, times 100. Each response counts once, no matter how many of your URLs it cites, which keeps the number readable as 'the share of answers that use us as a source'.
Citation rate is easy to confuse with its neighbors, and reporting one in place of another misleads:
- Citation share is citation-level: your domain's portion of all citations recorded, so it depends on how many other sources each answer lists.
- Cited URL rate is the same idea at page level: the share of responses citing one specific URL.
- Mention rate counts answers that name your brand in the text, whether or not they link to you. An answer can cite your page without recommending you, or recommend you without citing you.
- Self-citation rate is usually defined as your own domain's share of all citations, which makes it a citation share rather than a citation rate, despite the name.
The denominator needs care. Citations only exist when a model searches the web, and many API models answer from training data and cite nothing. A blended rate across live-search engines and non-searching models drags the number down for reasons that have nothing to do with your content. Report it per engine, and say whether responses without web search are included.
The ceiling also differs per engine because answers carry different numbers of sources. Promptwatch's average sources per response data shows ChatGPT citing about five sources per web-search answer and Google AI Overviews and Perplexity about ten, so the same content can earn a higher citation rate on engines that list more sources. Platform changes move it too: ChatGPT's average dropped from about 6.4 to about 4.7 sources per answer after the GPT-5.3 rollout in March 2026, which lowered citation rates broadly without any brand changing its pages.
For GEO reporting, freeze the prompt set between periods, sample each prompt more than once per cycle (AI answers vary between runs, see AI answer volatility), and split by prompt type. Branded prompts usually cite the brand's own site often, while category prompts lean on third parties, so a shift in prompt mix can move the blended rate on its own. Compare against named competitors on the identical prompt set, and read citation rate together with mention rate: the gap between them shows whether your content is used as evidence or your brand is merely talked about. Our AI search visibility KPIs guide covers how to put these side by side in a report.
Examples of Citation Rate
- A brand cited in 18 of 200 tracked ChatGPT responses in a week reports a 9% ChatGPT citation rate for that week.
- A team excludes responses from models that did not search the web and sees its blended citation rate rise from 6% to 14%, then switches to reporting per engine.
- A fintech company finds a 40% citation rate on prompts naming its brand but 3% on category prompts, and focuses content work on the category set.
- An analyst notices citation rates for every brand in a category fell in the same week in March 2026 and traces it to a platform-wide drop in sources per ChatGPT answer.
