Citation velocity

What is citation velocity?

Citation velocity is the rate at which new AI-engine citations to a source accumulate over time. Where attribution rate is a point-in-time measure (what fraction of queries cite you now), citation velocity is its time-derivative: new citations per fixed window, by engine, for a given query set. The basic formulas:

Citation Velocity (period N) = new citation instances observed in period N (per window length)

Citation Acceleration (period N) = Velocity (period N) − Velocity (period N−1)

Velocity is the raw count of citations new to period N (units: citations per week, citations per month). Acceleration is the change in velocity period over period (units: citations per week vs the prior week). The two are routinely conflated in vendor blog framings: treat them as distinct metrics here. Most "velocity tracking" practice tracks velocity directly; acceleration is the second-order derivative for spotting trend inflections.

Citation velocity is a practitioner-conceptualized GEO metric. The term itself is borrowed from academic bibliometrics, where Garfield-era citation analysis (1955 onward) established the citation-counting infrastructure that later enabled rate measurement as a standard scholarly metric in the Scientometrics journal and adjacent literature¹. The same name applied to AI-engine citations is a recent extension. Several vendor blogs have published "citation velocity" definitions for AI search (UltraScout's Citation Velocity Index, Rankeo's Citation Velocity Score, Steakhouse's Citation-Velocity Standard), typically as single-window or competitor-relative measures². No peer-reviewed academic literature defines the AI-engine version of the metric, and none of the major GEO tracking tools (Profound, Otterly, AthenaHQ, Peec, Brand Radar) currently exposes citation velocity as a named field in their public-facing dashboards. The per-engine + per-query-set + novelty-storage operational discipline described below is more rigorous than the published vendor blog framings, and is the contribution this entry adds to the citation-measurement cluster vocabulary.

Three operational choices the practitioner must lock before reporting a number:

• Window length. Weekly is the minimum useful granularity (AI engine indexing lag is typically days to weeks). Monthly is reasonable for executive reporting but obscures short-lived index refresh events. Comparisons across different window lengths are not directly meaningful.

• Novelty definition. Track five distinct citation states per (source, engine, query-cluster, period), not just a binary new vs persistent split:

  • first-seen: never appeared in any prior period's probe list
  • new-vs-previous-window: appeared this period but not in immediately prior period (default "new" for velocity reporting)
  • recovered: appeared in some prior period, absent in immediately prior, reappeared this period
  • persistent: appeared in both this period and immediately prior period
  • lost: appeared in immediately prior period but absent this period

  Default "citation velocity" counts new-vs-previous-window; reporting first-seen separately distinguishes true growth from recovered churn (an entry returning to the cited list after temporary absence is not the same as a fresh first appearance). This requires storing the full per-probe citation list across periods, not just aggregate counts.

• Engine breakout. Report per-engine velocity, not just aggregate. Engines have different ingestion lags (Perplexity often days; ChatGPT search and AI Overview often weeks); a cross-engine aggregate hides single-engine ingestion spikes that look like general velocity but are not.

Status in 2026

Emerging metric. Vendor blogs (UltraScout, Rankeo, Steakhouse) have started publishing "citation velocity" definitions, but no major GEO tracking tool currently exposes it as a named field in a public-facing dashboard. Profound, Otterly, AthenaHQ, Peec, and Brand Radar all surface point-in-time citation counts and weekly trend lines; whether their internal data models compute a velocity number explicitly varies and is generally not documented in their public marketing pages. Practitioners derive citation velocity manually by computing deltas between weekly probe rounds.

Why velocity matters distinctly from attribution rate: velocity is the leading indicator of brand-recognition shifts, where attribution rate is the lagging indicator of where you stand. A flat attribution rate over multiple weeks can mask both high-churn (new sources displacing old, signaling competitive pressure) and low-churn (stable repeats, signaling moat) realities. Velocity, computed from probe-to-probe deltas, separates the two. Practitioner experience suggests meaningful attribution-rate shifts may be preceded by 2 to 6 weeks of velocity acceleration that the rate itself smooths over (anecdotal practitioner range, not from controlled study). Treat this as a monitoring hypothesis, not a benchmark; whether the lead time holds across query types and engines has not been empirically established in published research.

Note on this entry's territory (paired with the attribution rate entry as its time-derivative; paired with the citation share entry as the temporal-vs-positional pair): the underlying concept (rate of new references over time) is academic-canonical in bibliometrics. The application to AI-engine citations with per-engine, per-query-set discipline is non-vendor-canonical: no vendor publishes this operationalization. The content-side application (writing in ways that survive AI engine index refresh) sits in practitioner-discipline territory, because content quality is something publishers can directly audit without needing vendor-confirmed ingestion mechanisms.

How to apply

Citation velocity is a measurement discipline, not a content lever. Three operational moves to track it cleanly:

• Pin a fixed query set and a fixed engine list. Velocity comparisons across time require everything else stationary. Lock 20 to 30 queries (the same set the attribution rate entry recommends) plus the engines you probe. Drop a query only when it becomes obsolete (vendor renamed a product, the question stopped being asked); do not drift the query set to chase rising topics, which conflates velocity with selection bias.
• Distinguish velocity from acceleration. If Week 1 has 3 new citations and Week 2 has 5 new citations, Week 2 citation velocity is 5 new citations per week (the raw count of citations new to that period); Week 2 acceleration is +2 per week relative to Week 1 (change in velocity). The two are routinely conflated in vendor blog framings; pick which metric your report tracks and label it accordingly. Store the full per-probe citation list across periods, not just aggregate counts, so you can compute both the velocity-vs-acceleration distinction and the new-vs-recovered-vs-persistent typology above.
• Watch the per-engine breakdown. Aggregate velocity hides single-engine ingestion spikes. If Perplexity newly cites you 4 times in Week 2 while ChatGPT search and AI Overview stay flat, the spike is Perplexity-specific (likely an index refresh or grounding-mode change), not a general velocity boost. Per-engine velocity tells you which engines are moving and which are stable.

What citation velocity does not measure: citation depth (whether you are primary or secondary in the cited list, captured by citation share), engine-traffic-weighted value (a citation in a high-traffic engine matters more for downstream traffic), or sustained quality (a velocity spike that decays in 2 weeks is not the same as a sustained level shift). Velocity is one axis of a multi-axis measurement program; reporting it alone over-rotates on short-term momentum.

How it relates to other concepts

• Time-derivative of attribution rate. Combine attribution rate (level) with citation velocity (trend) for a complete picture; a high attribution rate with negative velocity is a different story from a low attribution rate with positive velocity.
• Temporal-vs-positional pair with citation share. Citation share is positional (you vs others at one point); citation velocity is temporal (how fast you change over time). A flat citation share with high citation velocity means you and competitors are all moving up together; a flat citation share with low citation velocity means the field is stable.
• Distinct from citation match rate, which counts linked-vs-unlinked references at a point in time; citation velocity can be computed for matched citations only or for all attributed references, depending on what you are tracking.
• Related to cite-ability as outcome to property. Cite-ability is a property of content at the passage level; citation velocity is a measurement outcome at the program level. High cite-ability content may be associated with stronger citation velocity gains after AI engine re-indexing, but velocity is also shaped by engine ingestion lag, query-set scope, authority, and competitor citation churn; the direct causal link from passage-level cite-ability to program-level velocity has not been empirically isolated.
• Conceptual sibling to the academic bibliometrics term¹; same metric shape, different citation surface.