Machine Relations uses Citation Velocity for the change in observed citation presence between comparable collection periods.
Machine Relations definition: Citation Velocity records how observed citation presence changes between comparable collection periods for a fixed query-and-engine panel. It is the change measure paired with Citation Decay.
Premise: a defensible Citation Velocity series keeps the query set, engine set, collection method, and observation window fixed across comparable periods. A protocol change creates a new measurement basis and should be disclosed rather than blended into the prior series.
The measure can be reported in two forms:
Arithmetic illustration (not observed data): if a fixed panel gains twelve citation appearances and loses four between two collection periods, gross additions are twelve and net change is eight. The arithmetic describes the metric; it is not a benchmark or a claim about expected performance.
A Stacker/Scrunch case study reported that distributing the same story across third-party outlets raised the measured AI citation rate from 8% to 34% in its test (Stacker, 2025).
Research across six generative engines reported a 17.3% visibility improvement from structural optimization in the study's measured setting (Yu et al., 2026).
Liu, Zhang, and Liang evaluated citation correctness by testing whether cited passages entailed the claims attached to them in model-generated answers (2023).
Google states that its AI features can use query fan-out and supporting links from multiple web pages (Google Search Central).
Premise: these findings establish bounded observations about distribution, document structure, citation support, and retrieval. They do not establish a universal cause of Citation Velocity or a guaranteed outcome for any brand.
Premise: each observation should identify the query, engine, collection time, answer, cited source, and entity matched to that citation. The same matching rule should be used in every comparable period.
Premise: report additions and losses separately before calculating net change. A positive net value means additions exceeded losses in that measured panel; a negative value means losses exceeded additions. Neither result explains why the change occurred.
Premise: compare a brand only against competitors observed with the same instrument and collection window. Machine Relations has not established a cross-industry Citation Velocity benchmark.
Premise: Citation Velocity is descriptive. It records movement in observed citations; it does not prove that publishing, earned media, content structure, freshness, or any other input caused that movement.
Premise: read Citation Velocity beside the current level of observed citation presence. A change measure without its starting level can conceal whether the measured panel is large, small, stable, or volatile.
Premise: keep citation position, mention frequency, click-through, and business outcomes as separate fields. Citation Velocity cannot substitute for those measures.
Premise: Citation Velocity is separate from content volume, citation position, traffic, revenue, and buyer intent. It does not assign a cause to a citation gain or loss, and values from panels with different queries, engines, matching rules, or collection windows should not be treated as comparable.
Premise: the MR Stack places Citation Velocity in its measurement and feedback layer beside Share of Citation, Citation Decay, Entity Resolution Rate, and AI Visibility Score.
Working model (not measured): use a change in Citation Velocity as a prompt to inspect the underlying observations. Investigate additions and losses by query, engine, source, and entity match before assigning a cause or changing an upstream program.
Measurement method
Premise: compare two collection periods from the same fixed panel. Count citation appearances present in the current period but absent from the comparison period, then report losses separately and calculate net change if needed.
Setting a target
Premise: Machine Relations has not established a universal target. Set an internal comparison basis from a declared panel and observation window; do not treat it as a cross-category benchmark.
Relationship to earned-media timing
Working model (not measured): record earned-media publication dates as candidate explanatory events, then compare them with later citation observations. Timing alone does not establish that coverage caused a change.
Negative values
Machine Relations definition: Net Citation Velocity is gross additions minus losses for one fixed panel.
Comparison with AI Share of Voice
Premise: AI Share of Voice summarizes observed brand presence within a comparison set. Citation Velocity records change in observed citation presence between comparable periods. They describe different fields and should not be substituted for each other.
Machine Relations' own methodology, dataset, and research pages related to this term. These are self-references, listed separately from Sources — they are not independent evidence.
Related concepts
Supporting research
Framework context