Definition
Citation probability is the likelihood that a specific page or domain is cited as a source when an AI answer engine responds to a given prompt. Because AI answers vary between runs, it cannot be read from a single response. It is estimated as the share of repeated runs of the same prompt, on the same engine and in the same market, in which the source appears.
The term sits next to two observed metrics. Citation rate reports how often you were actually cited across a prompt set in a period. Citation share reports what portion of all citations you hold. Citation probability is the per-prompt, per-source estimate behind those numbers, and teams use it to forecast and prioritize: which prompts you are close to winning, and which you have almost no chance on today.
A page only gets cited if it clears several stages, and each one can push the probability toward zero:
- Access: the engine's crawlers and fetchers can reach the page (no robots.txt or CDN block on agents such as OAI-SearchBot, Claude-User or PerplexityBot).
- Retrieval: the page ranks in the search source the engine uses for the query fan-out sub-queries. Google's AI features documentation says a page must be indexed and eligible for a snippet to appear in AI Overviews or AI Mode, and TechCrunch reported in March 2025 that Claude's web search appeared to use Brave Search.
- Selection: the page wins one of a limited number of slots. Promptwatch's average sources per response data shows ChatGPT citing about five sources per web-search answer, against roughly ten for Google AI Overviews and Perplexity.
- Retention: the page keeps being cited as fresher sources appear. The same tracking data shows that half of cited news pages dropped out within 36 days, while landing pages and documentation held on for close to 90 days or longer.
Some factors are platform-wide and outside your control. When ChatGPT's average fell from about 6.4 to about 4.7 sources per answer after the GPT-5.3 rollout on March 4, 2026, every brand's citation probability on ChatGPT dropped at once. Separate those shifts from changes caused by your own content before reacting.
To estimate citation probability in practice, run each tracked prompt several times per engine on a fixed schedule, record whether your domain or URL is cited in each run, and calculate the proportion. With 10 runs, a page cited 3 times has an estimated probability of 30%, but the uncertainty is wide, so look at trends across weeks rather than single readings. Group prompts by probability band: near zero usually points to access or retrieval problems; 20 to 60% means you are a contender and content improvements can tip it; above 80% is a position to defend with content freshness.
What raises the probability is mostly practical: a page that answers the prompt's exact question near the top, current facts and dates, crawlable HTML, and coverage on the third-party pages that engines already cite for the topic. See our guide on how to grow citations in AI for the levers in more detail.
Examples of Citation Probability
- A SaaS team runs 'best invoicing software for freelancers' 12 times a week in ChatGPT and finds its comparison page cited in 4 runs, an estimated 33% citation probability worth improving.
- A retailer sees near-zero citation probability on Claude for all prompts, checks its CDN logs, and finds Claude-User requests returning 403 errors.
- A publisher notices its news explainers fall from roughly 70% to 20% citation probability within six weeks and schedules updates before the decline instead of after it.
- After a platform update cuts the number of sources per answer, a GEO team confirms that competitors' probabilities fell by a similar amount before rewriting any pages.
