I rewrote a page title for clicks and only later found it was my most AI cited page

Two weeks of Search Console data pushed me into a CTR experiment. Ranking was not my problem. Position improved from 41.1 to 27.6 over a month and the clicks did not follow, so I went after titles instead.

The first page I picked was a manufacturing cost post sitting at 1,342 impressions, 0 clicks, position 6.2. I retitled it to lead with the 4k to 30k price bands the page already contained. That one I still feel good about.

The second is the one I got wrong. I changed it from “Nearshore SaaS development: what the timezone overlap is actually worth” to “Nearshore SaaS development: 4 to 8 week builds, and when offshore wins”. Cleaner, leads with a number, a better title by every rule I was using.

The next day an unrelated check told me that same page is the single most retrieved page I have in generative answers, and that Perplexity had it as source 1 of 10 on a timezone overlap query the day before. I had just deleted the exact phrase it matches on from both the title and the H1.

I am not claiming I broke it. Perplexity is still answering from a cached copy that predates the edit, so the real test has not run yet. What bothers me is my own experiment log. It lists CTR, clicks and position as the review criteria and does not mention retrieval anywhere. If the number had moved I would have read a green result and never looked.

When you run an experiment on one metric, how do you decide which other metrics are allowed to move?