What is a longitudinal study?

A longitudinal study tracks one set of prompts, pages, or brands across repeated measurements, week over week or month over month, rather than checking once and stopping. The value of doing this is almost entirely about separating signal from noise: any single citation-rate reading carries sampling variance, but a trend line across ten readings makes a real, sustained shift visible against that noise in a way one reading never can.

The trade-off is cost and patience. A longitudinal study takes longer to say anything and requires the measurement method to stay the same across every round, same basket, same prompt phrasing, same model where possible, because changing the method mid-study makes it impossible to tell whether a shift in the numbers came from reality or from measuring differently.

This is also the setup that exposes model drift: a longitudinal study run against a fixed, unchanged page will still show movement over time, because the models being queried keep changing underneath it. That movement is real data, just not data about your content.