Responsive search ads: pin headlines or let the system rotate them?

Pinning gets a lot of hate in responsive search ads, mostly because it cuts down the combinations the system can test. I think it makes sense when you have a real reason, like a legal line you have to show or a brand name that must sit in position one. Pinning everything out of habit is where it goes wrong. The ad turns nearly static, the system can't rotate much, and you end up judging it on a sample too small to mean anything. For a campaign with a few thousand impressions a month, I'd pin one headline at most and leave the rest free. I'm less sure about high-volume accounts with lots of assets, where pinning two or three might not cost much. Has anyone seen a clear CTR or conversion drop after pinning, or is it mostly theory?

Asset labels (Low, Good, Best) only rank an asset against others in the same position. A headline pinned to position 1 gets compared with the other position 1 candidates, so it can sit at "Good" and tell you nothing about how it would do in a free slot. If you pin and then read the labels, you're partly measuring the pin.

For a dental clinic with modest volume, I'd keep the brand headline pinned, leave the rest free, and test with a proper experiment split instead of comparing the ad before and after. Before and after also picks up seasonality and bid changes, and at a few thousand impressions a month those will swamp any pinning effect.

Agreed on the split test, but start by checking how many conversions the clinic gets in a month, because that decides whether the test is worth running at all. With only a few dozen, a split can take months to separate two ad versions, and by then the seasonality you mentioned has already shifted. In that case I'd go by CTR and cost per click as the early signal and only trust conversions once the sample is big enough to mean something.

CTR has the same pin problem, though. A headline sitting in position 1 picks up more impressions and clicks no matter what the copy says. My rule of thumb is that if the test can't get to roughly 50 conversions per variant, only compare CTR between ads that share the same pins, and don't call a winner on cost per lead alone.

Agreed on the pin problem, and 50 conversions per variant is a sensible floor. The catch is that with only a few dozen conversions a month, you'll almost never hit that in a reasonable window, so the test mostly shows you noise. For a clinic that size I'd rather run it longer, or lean on a bigger lead-quality signal, than call a winner early.

For a clinic the form fill is the weakest number in the funnel. An ad that pulls curious people will win on enquiries and lose on bookings. If the booking system can send back which leads came in and which actually attended, import those as offline conversions tied to the Google click ID. That way the test runs on something close to real patients, not form submissions. Keep the form fill as a secondary metric so you still get volume while the sample builds.

Offline conversions are the only way I'd trust a clinic test too, but the click ID is where it usually breaks. Google can only match an attended booking back to the click if the gclid was saved with the lead, and clinic sites lose it a lot. Booking widgets often run in a third-party iframe, or the form redirects to another domain, and the URL parameters vanish along the way.

So before the test starts, submit a real booking from a paid click and check that the gclid shows up in the CRM record. If it doesn't, you'll be importing conversions that can't be matched, and the test will look like it's working when it's running on nothing.

Make sure attendance data flows back on a regular schedule too. Google limits how old a click can be when you upload a conversion for it, so a quarterly export can quietly drop most of your matches.

One test booking only proves the gclid can survive once, and drop-off usually varies by device and browser. A widget might keep the parameter on desktop and lose it on mobile. I'd watch the match rate in the offline conversion diagnostics over the first few weeks instead of trusting a single booking.