Here's the thing that reframes this whole thread: 10 to 3 conversions isn't a 75% performance collapse, it's a sample-size problem wearing a costume.
At around 10 conversions a week, normal variance alone will swing you 40-50% in either direction. A week with only 3 conversions has a confidence interval so wide it still includes your old average. Nothing gets restructured until there are at least 30 conversions post-change, until then, you're reading noise.
That said, two real suspects are worth checking, in this order:
🎯 The hospital-name negatives. those terms converted, and blocking them on the theory they weren't legit is backwards logic. Converting queries are customers until lead-quality data proves otherwise. A hospital name in a query is often just how people search for care nearby, or caregivers researching on behalf of a relative. Pull the search terms report for the weeks before the negative went live and check the conversion rate on those terms. If they hit your target CPA, that negative is costing money, not saving it. Golden rule: never negative a converting search term on vibes. Negative it on data (lead quality scores, downstream conversion rate).
🔁 The paused duplicates. With exact match, duplicate keywords across ad groups enter an internal auction and Google only serves one, so the others were barely spending anyway. Unpausing all of them just re-creates the auction chaos. Better move: open the search terms report for the 30 days before the change, find which keyword actually matched the converting queries, and keep exactly that one. Delete the real duplicates so they can't fight each other again.
Unchanged CTR basically clears the ad copy change, by the way. Start with the search terms report, it's the autopsy report.
rule of thumb for low conversion volume: change one thing at a time, then wait for 30 conversions before judging anything.
