Guide
Most teams treat placements as an export setting — render the master, let the platform reframe it, move on. That's how a good ad becomes a bad one without anyone editing a single frame.
The temptation is to think of placements as aspect ratios: 9:16 here, 4:5 there, 1:1 somewhere else. But the ratio is downstream of something more important — what the viewer agreed to when they opened the surface. Someone in a full-screen vertical feed has handed over the entire screen and expects to be entertained immediately; there is no other content competing for the frame, and nothing to look at if your ad is boring. Someone scrolling a mixed feed is skimming a list, sees your ad at partial size among posts from friends, and is deciding in a fraction of a second whether to stop. Someone tapping through Stories is in a rhythm of dismissal — their thumb is already moving before your first frame renders.
Same file, three completely different jobs. That's why the crop question and the aspect ratio question aren't the same question. Picking 9:16 is a decision about pixels. Deciding whether your ad survives a surface where the viewer is actively trying to skip it is a decision about the ad.
Automatic placements plus one master file is the standard setup, and it isn't wrong — it's just unsupervised. The platform will happily take your carefully composed 4:5 video and center-crop it to vertical, which means the product you framed on the right third is now half out of frame. Your text overlay, which sat in clean space in the master, is now under a profile bubble or behind the CTA button. Your opening shot, which read fine at feed size, is suddenly enormous and reveals that the lighting was mediocre.
None of that shows up as a creative problem in your reporting. It shows up as "Stories underperforms for us," which then becomes a placement exclusion, which then becomes a permanent belief about a surface you never actually tested. The ad wasn't bad on Stories. The version of the ad that ran on Stories was bad, and nobody ever watched it. This is the same class of oversight covered in the mobile-first ad design checklist: the thing that breaks the ad is almost never the thing you were looking at while making it.
You don't need a bespoke edit for every surface — that's how a two-day shoot turns into a two-week export marathon. Spend the effort where the budget goes. In practice that means: a purpose-built vertical cut for full-screen surfaces (the ones that usually eat the majority of short-form spend), a properly composed feed cut, and an honest check on everything else. If a placement is taking under a few percent of delivery, an auto-crop that you've at least looked at is fine.
What changes between cuts is more than framing. Full-screen placements tolerate — and reward — a slower reveal, because the viewer has already committed the whole screen. Feed placements need the payoff compressed forward, because you're competing with the next post rather than with nothing. And any surface with heavy interface chrome needs your text overlays re-placed rather than re-scaled. The cut is an edit, not an export preset.
The instinct is to split placements into their own ad sets and let the data decide. On a large budget, fine. On a normal budget, you've just divided your results across three surfaces so that none of them reaches significance — the same arithmetic problem behind how many ad variants to test at once. Testing placements is testing variants, and the variants multiply fast.
A cheaper sequence: export the exact file you would ship for each placement — not the master, the shipped cut — and screen each one before you spend. PreTestAds predicts how a given cut holds attention across its runtime and ranks it against a benchmark of top-performing short-form ads, so a crop that broke the composition tends to show up as a weaker read than the version it came from. That comparison is the useful signal: not "is this ad good," but "did this placement's version lose something the original had." Ship the cuts that hold, rework the ones that don't, and let live delivery answer the question it's actually good at — which surface buys attention cheapest.
Be clear about the boundary. Screening cuts tells you whether a specific file holds attention. It doesn't tell you which placement is cheapest for your audience, how the auction will treat you, or whether the offer converts once someone taps through — those are delivery and downstream questions, and pre-testing is a screen that complements live testing rather than replacing it. What it does buy you is the elimination of a stupid, common, entirely avoidable failure: paying to distribute a version of your ad that was broken by a crop nobody watched before launch. Check the cuts, then let the platform do its job.
Upload the placement export and see whether the crop cost you attention — first analysis free.
Score Your AdOnly if you've checked what the crop does to it. Automatic placements will take one master file and reframe it for feed, Stories, and Reels, which can push your subject off-center, bury text under interface elements, or turn a composed shot into an awkward one. The safest default is a dedicated cut for the placements carrying most of your budget, and a checked fallback for the rest.
Split placements into separate ad sets and you'll fragment your budget across surfaces that each need their own results before they mean anything. A cheaper sequence: export the actual cut you'd ship for each placement, screen those exports for predicted attention before launch, ship the ones that survive, and let live delivery sort the rest. You spend on the versions that already look viable rather than paying to discover a bad crop.
No. Attention scoring reads the creative file itself, not the auction, the audience, or the placement's cost. It can tell you whether a specific export holds attention — which catches the common failure where a crop quietly breaks an ad that worked fine in its original framing. Which placement is cheapest for your audience is a delivery question that only live data answers.