Measured engine
ChatGPT
What we measure about whether ChatGPT cites your site in its answers, how each sample is a real query, and the limits of that number.
Citation sampling on ChatGPT unlocks on the Pro plan and up, because each sample is a real paid query. See pricing.
What we measure on ChatGPT
- Answer Share. A fixed prompt set is run against ChatGPT, several samples per prompt, with the strict-majority rule deciding whether a prompt counts as citing you.
- Citation Rate and Owned Citation Share for ChatGPT citations across the window.
- Average citation position where ChatGPT exposes a source order.
- Ghost citations, where an answer draws on your page without naming your brand.
- Run over run change, gated by the same interval-plus-Fisher test used on every engine.
How we sample it
Each ChatGPT sample is a real per-query call to OpenAI, so we sample deliberately and report the count. Every share ships with a 95% Wilson interval and its sample size, labeled provisional below three samples per prompt. A crashed run writes no aggregate, because a missing measurement is a gap and not a zero.
What we do not measure here
Stating the limit is the point. A number is only trustworthy next to the boundary of what it does not cover.
- Referral traffic from ChatGPT. We measure whether ChatGPT cites your domain, not sessions it sent.
- Whatever ChatGPT learned in training but does not surface as a citation. The number is about cited sources in the sampled answers, not model memory.
- Any engine outside this one. A ChatGPT result is labeled ChatGPT and never presented as an all-engine figure.
The full method, with every denominator and error treatment, lives on how we measure. When another tool shows the vocabulary without the math, hold it up to measured vs asserted.
Every surface we measure
Engines we sample
Crawler families we observe