Citation probability is the percentage of repeated AI-search responses in which a specified source or domain is cited for a defined prompt or prompt set. It is a metric defined by blimpp, published here with its full measurement method. It exists because AI engines do not return the same answer twice: a source cited in one response may be absent from the next, so a single screenshot proves very little about visibility.
In one sentence
Citation probability tells you how reliably an engine reaches for a given source when asked the same question repeatedly, rather than whether it happened to cite it once.
How citation probability is calculated
Citation probability = (responses citing the source ÷ total responses tested) × 100
The denominator is the total number of responses collected for a defined prompt or prompt set on a single engine under controlled conditions. The numerator is the count of those responses in which the source appears as a linked citation, not merely a text mention. Measurement conditions follow the repeated-run protocol, and what counts as a citation is set out in measuring citations, mentions and recommendations.
Worked example
A prompt is run 100 times on one engine in a fixed period. The target domain appears as a cited source in 27 of the 100 responses. Its citation probability for that prompt is 27%. Illustrative example: blimpp benchmark data is added to this page as studies publish.
Why citation probability matters
Being cited is the mechanism by which a domain earns presence inside AI answers, and probability is the honest way to express it. A brand reporting “we are cited by ChatGPT” on the strength of one response may have a citation probability of 5% or 85%; the two situations demand completely different action. Probability also makes progress measurable: an intervention either moves the number across repeated runs or it does not.
What affects citation probability
Evidenced factors: whether the engine’s crawler can access and index the page at all, since blocked or unindexed content cannot be cited; and whether the page contains self-contained, extractable passages that answer the prompt directly, which platform guidance from Google and OpenAI consistently emphasises. Factors with strong practitioner evidence but less formal documentation: freshness of the page relative to the query, the domain’s presence across independent third-party sources, and how closely the page’s stated topic matches the sub-queries the engine generates. These are treated as working hypotheses until tested.
Related concepts
References
- Google Search Central, Optimising your website for generative AI features on Google Search: developers.google.com
- OpenAI Help Center, Publishers and Developers FAQ: help.openai.com
Author: Harpal Singh · Last reviewed: 7 August 2026