AI Citation Metrics, Defined: Rate, Frequency, Velocity, and Share

Five different numbers all get called "AI citation metrics," and teams routinely track the wrong one for the question they are asking. Citation rate answers "am I visible." Frequency answers "how deeply does the engine rely on me." Velocity answers "am I improving." Share of citation answers "who owns this category's answers." Share of model answers "does the model even know my brand." This page defines each one with its formula and when to use it, and links to the deep-dive for the metrics that have one. All five sit on the measurement layer of the AI citability framework: they quantify the outcome that the framework's five pillars exist to produce.

What Is AI Citation Rate?#

Citation rate is the share of test queries where an AI engine cites your domain as a source.

citation_rate = (responses citing your domain / total responses tested) x 100
# measured per platform, against a fixed query panel

It is binary per response: one response either cites you or it does not, no matter how many of your pages it links. Rate is the north-star metric because it is normalized (comparable across time and across sites using the same panel size) and because it directly proxies the thing you want: presence in the answer. The full measurement protocol, including panel design and per-platform gotchas, is in How to Measure Your AI Citation Rate, and real numbers from six audited domains are in Benchmarking AI Visibility Across 6 Sites.

What Is AI Citation Frequency?#

Citation frequency is the raw count of citations your domain earns across a response set, counting every citation, including several inside one response.

citation_frequency = total citations to your domain across the response set
# a response citing 3 of your pages: rate counts 1, frequency counts 3

Frequency is the depth metric. Two sites can share a 10% citation rate while one earns a single link per answer and the other earns three: the engine relies on the second site more, and frequency is the only metric that shows it. It is also the metric you get without running a panel at all: Bing Webmaster Tools' AI Performance tab reports Copilot citation counts for verified sites, which makes frequency the cheapest continuous signal available today. Its weakness is the flip side of its strength: it is unnormalized, so it inflates with panel size and cannot be compared across different query sets.

What Is AI Citation Velocity?#

Citation velocity is the rate of change of your citation rate between two same-panel measurements.

citation_velocity = (rate_now - rate_previous) / interval
# expressed in percentage points per quarter; the panel must stay frozen

Velocity is the trajectory metric, and for newer sites it is more informative than rate itself: rate lags structural fixes by one or two crawl and index cycles, so a low rate with strong positive velocity means the fixes are compounding. The formula rules, a real 90-day worked example, and the calibration table for reading positive and negative swings are in AI Citation Velocity: Formula, Benchmarks, and a 90-Day Example.

What Is Share of Citation?#

Share of citation is your slice of all citations in the response set: the competitive view.

share_of_citation = (citations to your domain / total citations in response set) x 100
# 20 of 400 total citations = 5% share

AI answers are zero-sum in a way search rankings are not: a typical response cites 3 to 6 sources, and every citation you gain displaces someone. Share of citation is the metric for category-ownership questions ("of all the answers in our space, what fraction route through us?") and the one to put in front of an executive who thinks in market share. The term is young; it appears in the wild as both "share of citation" and "citation share" from early trackers like AuthorityTech, and no measurement standard exists yet, so state your panel and counting rule whenever you report it.

What Is Share of Model?#

Share of model is the fraction of category answers that mention your brand at all, linked or not, versus competitors.

share_of_model = (responses mentioning your brand / total category responses) x 100
# counts unlinked mentions; citation not required

This is the AI-era descendant of share of voice, and it measures something the citation metrics structurally miss: recognition from training data. A model can recommend a brand it cannot cite, and it can cite a domain for a brand it barely recognizes. The gap between share of model and citation rate is itself diagnostic. A healthy share of model with a near-zero citation rate means the model knows you but cannot retrieve a citable page, which is an infrastructure problem, the exact failure mode dissected in Why Your Brand Is Mentioned in ChatGPT but Not Linked. The inverse (cited but never named unprompted) is an entity-authority problem.

Which Metric Should You Track First?#

Track them in the order they become measurable:

Every one of these assumes the same precondition: a site whose infrastructure does not cap the numbers at zero. Before investing in a measurement cadence, run the free scan to check the 10 structural signals that block citation regardless of content quality. For the category-level context on why answer engines reward measurable, structured sites in the first place, see Answer Engine Optimization Explained on chudi.dev.