What is a control group?

Say you rewrite a product page and, two weeks later, its citation rate is up. Was it the rewrite? Or did the underlying model get updated in that window, or did the topic get more attention generally, or did your prompt basket just land differently that week? Without a control group, pages or prompts you deliberately left alone, there's no way to separate your change from everything else moving at the same time.

In AI visibility work, the biggest uncontrolled variable is the model itself. Providers update models and retrieval systems on their own schedule, and those updates shift citation patterns across the board, unrelated to anything a given brand did. A control group of untouched pages, tracked over the same window as the pages you edited, is what lets you attribute a change to your edit rather than to a shift happening underneath everyone at once.

This is easy to skip because it doubles the measurement work, you're tracking two sets instead of one. But a before-and-after number on a single changed page, with no control, is a story, not evidence.