Advantage+ creative enhancements: let Meta edit your ads or keep the originals?

Meta's automatic creative enhancements (cropping, brightness, generated text variations) are on by default in a lot of accounts now. For anything with a specific brand look or a founder on camera, I'd turn the generative text and visual tweaks off. The system will happily rewrite your headline into something that sounds like every other ad in the feed.

For cold prospecting, where you're testing hooks anyway, leaving the crop and format tweaks on is low risk and can help delivery across placements. The catch is you won't know which change helped unless you run the enhanced and original versions against each other. Has anyone split that test with enhancements on and off, and did conversions move or just CTR?

For e-commerce, CPA and CTR can both mislead on this. Rewritten headlines often pull curiosity clicks that look great in the dashboard but convert worse or land smaller baskets, so compare purchase rate and average order value between the arms too. Also make sure both versions got comparable spend, because delivery tends to drift toward one of them, and a lopsided test won't tell you much.

Agreed on purchase rate and AOV. A good rule of thumb is to hold off on calling a winner until both arms have spent the same share of budget over the same window. Delivery skew can mean one arm only reached the cheap, easy-to-convert audience, so a "win" is often just the arm that got better traffic.

Spend share isn't the thing to match. An arm can get the same money and still reach a different crowd if the test isn't a true randomized split, so budget parity doesn't fix audience skew. Meta's A/B test feature divides the audience before delivery, so check that setup first. If the split is clean, a cost per purchase gap that holds over a few weeks tells you more than whether spend landed within a few percent. Forcing spend parity on a test that isn't randomized just makes you wait longer for a result that's still unreliable.

Fair on the randomized split, but whether a few weeks of cost per purchase tells you anything depends on how many purchases each arm gets. If each arm only logs a handful of conversions a week, one good day can flip the gap, so you'd need a much longer window before trusting it.

Agreed on volume, and I'd put a rough number on it. If each arm has fewer than about 30 purchases in the window, I wouldn't trust cost per purchase at all. Say a store gets 8 purchases a week per arm. A 20% gap at that count is mostly noise. In that case, use add to cart as a secondary read, since it gives you far more events, and keep the test running until both arms have a few dozen purchases each.

If you're going to lean on add to cart, make sure both arms turn carts into purchases at a similar rate first. If one arm's carts convert much worse, the extra events just paper over the same problem. A simple rule keeps you honest: only trust ATC when cart-to-purchase is within a few points across arms.

Honestly, it depends on what the enhancement actually changed. If the arms only differ in crop or placement, matching cart-to-purchase is a fair check. But if the generated headline changed the promise, a curiosity hook can fill carts with people who expected something else, and they drop off at shipping or price. Cart-to-purchase can still look fine in that case, because the carts that do form are mostly the committed ones. So for text-level changes I'd treat add-to-cart as a warning sign only and keep waiting for purchases before calling anything.