Citation Velocity
Citation velocity is the rate at which your AI citations grow or decline over time. Learn how to calculate it and why it moves before your visibility does.
TL;DR
Citation velocity is the rate at which your citations in AI answers grow or decline over time. Where citation frequency is a level, velocity is the slope: measured this week against last, or this month against last quarter.
Momentum matters here because AI visibility compounds. A page cited once becomes a source engines return to, so the direction of travel tells you more about the next quarter than the current number does.
Key benefits of tracking citation velocity
Levels tell you where you stand. Velocity tells you where you are heading, which is the number that justifies continuing or stopping a piece of work.
- Early signal: Velocity turns before the level does, so a decline shows up weeks before the headline rate drops.
- Work attribution: A step change after a content push or a placement campaign is the closest thing to proof that the work landed.
- Competitive warning: A rival's rising velocity on prompts you own predicts displacement before it happens.
- Budget defense: A trend line survives a leadership review in a way a single-week figure does not.
How to calculate citation velocity
Fix your measurement window
Weekly suits most teams. Anything shorter mostly captures the run-to-run variance built into AI answers.
Hold the prompt set constant
Velocity compares two periods, so any change to the denominator between them invalidates the comparison. Version the prompt set and date every edit.
Compute the change
Subtract last period's citation rate from this period's, then express it as a percentage-point change or a percentage of the prior rate. State which one you used, because the two produce very different-looking numbers.
Smooth the series
Use a rolling average across three or four periods. Raw week-to-week movement in a non-deterministic system is mostly noise.
Set a significance threshold
Decide in advance how large a change counts. If the confidence ranges of two periods overlap, nothing happened.
Annotate the timeline
Mark content launches, placements, model updates, and site changes on the chart. Velocity without context tells you something moved and never why.
Citation velocity vs. citation frequency
Citation velocity: The rate of change between periods. It answers whether things are improving and how fast.
Citation frequency: The level at a point in time. It answers where you stand right now.
Read together they diagnose the situation: low frequency with high velocity is an early-stage program working, while high frequency with negative velocity is an established position eroding, which is the more urgent of the two.
Calculating this properly means holding a prompt set stable for months and sampling repeatedly to separate movement from noise. Findrix runs your market's prompts across seven engines every week, keeps the set versioned, and shows confidence ranges so a real move never gets confused with variance. Every gap comes with the fix already written: technical, content and off-site. The audit is free, takes about a minute, and requires no signup.
Metrics to read alongside citation velocity
- Source turnover: How much the cited domain list changes between periods, which often moves before your own rate does.
- Competitor velocity: The same calculation for named rivals on the identical prompt set.
- New-page time to first citation: How long a freshly published page takes to get cited, which measures how fast engines pick you up.
- Post-update delta: The size of a category-wide shift after a known model release, which separates external causes from your work.
- Velocity by prompt type: Movement on comparison and pricing prompts against definitional ones.
Why velocity readings mislead
Small denominators exaggerate everything. Moving from two citations to four is a hundred percent increase and almost certainly noise, which is why percentage-point change is safer than percentage change at low rates.
Model updates create step changes that look like results. A category-wide jump the week a new model ships is the engine changing its behavior, not your content earning citations, and reading it as success leads teams to double down on work that did nothing.
The subtlest trap is a quietly shifting prompt set. Someone adds prompts for a new product line, the rate moves, and every historical comparison silently breaks with nobody noticing for months.
Frequently asked questions
How often should I measure citation velocity?
Weekly measurement with a three or four period rolling average works for most teams. That is frequent enough to catch model updates and competitive moves within days, and smoothed enough that normal variance does not read as a trend.
What is a good citation velocity?
Positive and sustained beats a large one-off jump, since single spikes usually trace to model updates rather than your work. The more useful comparison is your velocity against named competitors on the same prompt set in the same period.
Why did my citation velocity go negative without anything changing on my site?
Usually something changed around you. A competitor published a stronger source, an engine updated its model, or a page that used to cite you was refreshed. Check source turnover first, since the cited domain list typically moves before your own rate does.