Stop Testing Hook Swaps: The 90/60/30-Second Script Method for Meta's Andromeda Era

Picture two creative batches landing in the same ad account on the same morning.

The first is the playbook most of us ran for years. One hero video, then five versions of it where the only thing that changes is the first three seconds. Same body, same offer, same everything, just a fresh hook bolted on the front. Five ads, one real idea.

The second batch is one idea too, but built differently. A 90-second cut, a 60-second cut, and a 30-second cut of the same angle. Different scripts, different hooks, genuinely different edits, all pointing at the same message.

A year ago I'd have backed the first batch to give you cleaner data. Today, in most accounts I look at, the second one is the one that actually spends. The first one quietly dies, and not because the hooks were bad.

Here's my take on why that flipped, and how we structure batches around it now.

What Andromeda actually changed

Meta's newer ranking system, the one everyone's calling Andromeda, did something specific that broke the old approach. It got much better at spotting when two ads are basically the same ad.

Read Meta's own framing of it and the language is telling: the system favours creative individuality. It is built to reward genuinely different creative and to stop you stretching the life of one ad by spinning out a dozen tiny variations of it. The thing we all leaned on, the micro-iteration, is exactly the thing it's designed to flatten.

So when you upload five videos that share 90% of their footage and differ only in the opening line, the algorithm doesn't see five ads. It sees one ad with five hats. One or two of them get the impressions, the other three or four barely get out of the gate. Not because Meta is being stingy. Because as far as the similarity scoring is concerned, you handed it the same creative five times.

The tell that this is now formalised: Meta has been rolling out creative-similarity reporting, where you'll actually be able to see how alike two ads are inside the account. When the platform is willing to show you a similarity score, that's a fairly loud signal it's already using one to decide what spends.

Why hook swaps stopped earning their keep

I want to be fair to the old method, because it wasn't stupid. For years, swapping hooks on a proven body was the highest-impact thing you could do. The body was doing the selling, the hook was doing the stopping, and testing five openers on a winning base was a sensible way to find the one that stopped the most people.

The logic still holds for a winning ad. The problem is the supply.

When the algorithm collapses your five near-identical variants down to one or two that actually get spend, you've spent the editor's whole afternoon producing volume the auction won't even look at. Your hit rate per batch drops, your data comes back thinner, and you're left thinking your creative is tired when really you just fed Meta the same thing five ways and it picked a favourite.

In reality, the constraint moved. It used to be "find the best hook for this body". Now it's "give the algorithm enough genuine difference that more than one ad in the batch is allowed to compete".

The 90/60/30 build

This is the structure I keep coming back to, and it's the cleanest answer I've seen to the similarity problem.

Take one angle. Not five angles, one. Then build it at three lengths:

• A 90-second cut. Room to set up the problem, twist the knife a little, demo the product, land the offer. The long-form version for the people who'll actually sit through a story.
• A 60-second cut. Same angle, tighter. A different hook, a different lead, less setup, faster to the point. Not the 90 with the middle deleted, a genuinely re-scripted version.
• A 30-second cut. The punchy one. Different hook again, straight into the strongest line, in and out.

The key word is "different". Different scripts, different opening lines, and because the runtimes are so far apart, naturally different footage and pacing. To the similarity scoring, these read as three distinct creatives, so all three are allowed to compete. What I've watched happen is two, sometimes all three, start picking up spend in the same campaign, where the old five-hook batch would have force-fed everything to one ad.

And here's the bit that makes it workable for your editor: it isn't three times the effort. You're not writing three unrelated concepts. Same angle, same messaging spine, broadly the same clip library. The lines and the cut change, the research behind it doesn't. So the lift on the production side is modest while the thing Meta cares about, real creative difference, goes up sharply.

What this isn't

A couple of honest caveats, because I don't want anyone reading this as a magic formula.

This is not "length variants always beat hook tests". For a proven winner, iterating still has its place. Once an ad is genuinely working, going back in and improving the messaging, the imagery, the hold, that's still worth doing. The 90/60/30 idea is about how you structure a fresh test of an unproven angle, where the old instinct to spin five hooks off one body is the thing quietly hurting you.

It's also not a reason to chase length for its own sake. The point was never the seconds on the clock. The point is forcing three genuinely different executions of one idea so the algorithm treats them as three ads. Three lengths is just the most reliable way I've found to guarantee that difference without inventing three separate concepts.

And if your one angle is weak, three cuts of a weak angle is still a weak test. The research that picks the angle in the first place is doing more work than the format ever will. Format is how you give a good angle its best shot, not a substitute for having one.

Where the flexible-ad option fits

There's a related move worth knowing, for the cases where your variations really are tiny and you don't want to abandon them.

Meta's flexible ad format lets you load several pieces of media into a single ad unit and let the system serve whatever combination works best. The interesting use isn't "throw everything in and hope". It's that grouping near-identical iterations into one flexible unit can extend their life, because instead of five lookalike ads competing against each other and triggering the similarity penalty, you've got one unit quietly cycling through the variations.

So the way I'd frame the choice: if you've got genuinely different ideas, give them their own ads and let them compete, which is what the 90/60/30 build is for. If you've got a cluster of tiny tweaks, a different headline colour, a slightly different background, the same product on a new backdrop, those belong together in a flexible unit rather than fragmented into separate ads the algorithm will just pick one of.

Either way, the principle underneath is the same. Stop handing Meta five copies of one thing and acting surprised when it spends on one.

Where to from here

If your batches have been getting thinner lately, plenty of ads launched, only one or two ever earning real spend, it's worth checking whether your testing structure is the actual culprit before you blame creative fatigue. Nine times out of ten the angle was fine. The batch was just five hats on one head, and Andromeda only let one hat out the door.

Try restructuring your next test around it. One angle, three lengths, three honestly different scripts, and watch how many of them the auction lets compete. If two or three start spending where you'd normally see one, you've found the leak.

If you'd rather have someone go through your last few months of creative tests and tell you straight where the spend's been getting strangled, that's a big part of what we do in a Signal/Noise Audit. We'll pull apart how your batches are built, where the similarity scoring is quietly killing variants before they get a fair run, and which angles deserved more room than the structure ever gave them. No pitch attached, just a clear read on where your testing is leaking.