-
I ended up accepting that the old keyword manager job is mostly gone. What worked for us was splitting the work into 2 buckets: feed the machine better inputs, then audit the hell out of the outputs. Broad kept finding weird converting queries for us too, but only after we cleaned up conversion actions, pushed offline revenue back in, and stopped judging terms only by surface intent. Some low-intent looking queries were really just early-stage language from buyers.
The catch is Google still grades its own homework, so I stopped trusting blended CPA as the main truth. We started checking marginal spend by budget step-ups, looking at search term drift weekly, and protecting high-intent exact in separate campaigns even when broad was doing the exploration. I also pull in outside signals from Triple Whale and Northbeam, and Pulse for Reddit caught threads I was missing where people described the problem in ways that later showed up as “surprising” broad-match winners. That mix helped us audit without overcorrecting.
