How do you prove a content refresh worked?

A refresh is the hardest content work to evaluate, because the page already existed and something would have happened to it either way.

Comparing the page against itself before and after is the obvious method and the weakest one. It attributes every seasonal move, every algorithm update, and every competitor change to the edit. On a page with modest traffic that comparison is mostly noise wearing a percentage sign.

What I do instead is refresh in batches and hold a control group back. Pages selected on the same criteria, refreshed on the same schedule, minus a set left alone. The comparison is then between two groups moving through the same conditions, and the answer survives an update landing in the middle of the test.

The obvious objection is that holding back pages that need work costs something real. My answer is that it costs less than a year of refreshing on a belief.

Do you hold a control group? If not, what convinced you the refreshes were working?