AI ad copy: let it write from a blank page or critique your draft first?

I keep seeing people ask an AI to write a full set of Search ads from scratch, then wonder why half the headlines could belong to any competitor in the category. For most small accounts I'd flip the order. Write three or four rough lines yourself, then ask the AI to attack them. Give it the offer, the one thing a customer can't get anywhere else, and the search terms that actually converted last month. Then have it flag any claim you can't back up and any line that would fit a dozen other businesses. The blank-page version gives you volume, but the critique version keeps your facts and your voice in the ad. Curious how others split the work, especially if you have a lot of search terms data to feed in.

Agreed on the flip, but "search terms that actually converted" only means something once you name the conversion action. Plenty of accounts count form views, short calls, or GA4 imports that double-count, so the AI ends up praising terms that never led to a booking. And for a dental clinic with maybe six appointments last month, one term with two conversions is noise, not a pattern. I'd export the search terms report filtered to the primary conversion action, over a 30 to 60 day window, and feed that to the AI. That way the critique is working from real signal instead of a lucky month.

Jun's right about the conversion action, and the answer really depends on monthly volume. A clinic doing six bookings a month won't give you enough terms with real conversions even over a 60 day window, so use the critique pass for voice and claims, not for picking keywords.

I'd push back on keeping keywords out of the critique pass. Low volume makes the positive side weak, but the negative side still works. A term like "dental clinic jobs" or "free teeth cleaning" that got clicks and no bookings is an intent call, and the AI can make it from the query text alone without waiting for a pattern. So for a clinic with six bookings a month, I'd build the negative list from the search terms report now and save the positive keyword decisions for a longer window.

Agree on the negatives, Atlas. I'd cut right away when the query signals the wrong intent, like jobs or free. For ambiguous terms, wait until they've burned two to three times your target cost per booking with zero conversions, otherwise you end up blocking terms that just haven't had enough clicks yet.

Rowan's threshold makes sense, but how many clicks a term needs before "zero conversions" means anything depends on your conversion rate. If a clinic converts at 2% on clicks, a term with 40 clicks and no booking is still unclear. The same result at 10% would be a real signal. So check the click count against your conversion rate before you block anything.