Stop Splitting Your Meta Ad Sets by Platform

The creative advice still holds. The budget advice doesn't.

The standard line on Meta is that Instagram and Facebook need different creative, and that you should separate your ad sets by platform so you can see what each one is really delivering and move budget accordingly.

The first half of that is more true than it's ever been. The second half is now one of the more reliable ways to make an account underperform.

‍

Two apps, two models

In January, Meta published a fairly detailed account of what changed in its ads ranking systems through the end of 2025. Two things in it are worth an experienced reader's attention.

The first is that Meta doubled the GPUs used to train GEM, its ads ranking model, and moved to a sequence-learning architecture that reads longer stretches of user behaviour. Crucially, it now folds in additional organic engagement data on Instagram. That's the part most people skimmed past. The system deciding whether to show your ad on Instagram is learning from how that person behaves with organic Instagram content, which is a different behavioural fingerprint from the one it builds on Facebook.

The second is Meta Lattice, which consolidated Facebook Stories and other Facebook surfaces into a single Facebook model, while a separate run-time model runs across Instagram Feed, Stories and Reels. The reported outcomes differed by app: a 3.5% lift in ad clicks on Facebook, a bit over 1% gain in conversions on Instagram, and a 3% conversion rate improvement from the Instagram run-time model. Meta's own numbers, from Meta's own reporting, so read them as directional. But the shape is clear enough.

Instagram and Facebook aren't two placements inside one system. They're two systems that happen to share a campaign manager. Which is exactly why creative built for one rarely transfers cleanly to the other, and why "resize the asset and tick both boxes" has always been a weak strategy.

‍

Where the old advice breaks

Here's the problem. If creative is what the ranking model reads to decide who sees your ad, then every ad set you create divides that signal.

Split by platform and you've halved the learning data each side gets. Split by platform and audience and format and you've quartered it. You feel like you've gained control because you can now see a clean Instagram versus Facebook comparison in a report. What you've actually bought is a slower, thinner, worse-informed delivery system, and a set of numbers with less volume behind them than the ones you were trying to improve on.

The instinct is a decade old and it made sense when targeting was the lever. When you were picking interests and lookalikes, structure was how you expressed strategy. That's no longer where the decisions get made.

‍

The attribution problem underneath it

There's a specific claim in most versions of this article that deserves scrutiny: that Instagram spends most of the budget while Facebook delivers most of the conversions.

Plenty of people have observed that pattern. Fewer have asked whether it's real or whether it's an artefact of how the platform assigns credit.

Meta itself has effectively answered that. Its incremental attribution feature, which optimises towards conversions that wouldn't have happened anyway rather than conversions that merely happened after an impression, reported a 24% increase in incremental conversions against the standard attribution model in its Q4 rollout. Meta says it reached a multi-billion-dollar annual run-rate within seven months of launch, which tells you advertisers are finding a meaningful gap between the two views.

A 24% difference is not a rounding error. It's a different account. And if placement-level ROAS in Ads Manager is what you've been using to justify moving budget from Instagram to Facebook, you should know that the platform now sells a competing version of that number and thinks the standard one systematically undercounts.

Upper-funnel placements are exactly where last-touch attribution is weakest. Instagram Reels and Stories are exactly where upper-funnel work happens. Draw your own conclusion.

‍

So what should you actually do

Keep making platform-specific creative. That advice was right and remains right. Instagram rewards motion, pace and something worth stopping for. Facebook still rewards context, social proof and copy that gives someone a reason to click. These are different jobs and one asset rarely does both.

But stop expressing that difference through account structure. Put the variety inside the ad set rather than between ad sets. A consolidated ad set carrying genuinely different creative, different hooks, different formats, different lengths, gives the ranking model the range it needs to work out who each piece belongs in front of. That's what it's built to do now.

Your job has shifted from allocating budget between placements to supplying enough creative variance that the allocation is worth making. That's a production problem, not a media buying problem, and most teams are still staffed for the old one.

Two other things worth doing this quarter. Turn on incremental attribution and run your own comparison before you accept any placement-level conclusion you've been carrying around. And if you want to know what Instagram is genuinely contributing, don't read a report, run a holdout. Turn it off somewhere for four weeks and look at total revenue in your own system.

‍

The honest version

The old advice on this topic was built for a system where you decided who saw your ads. That system is gone. Meta reads your creative, models the person, and makes the call.

You still control two things that matter enormously: what you give it to read, and how you define a result worth optimising towards. Everything else is arranging deck chairs, and increasingly the deck chairs are bolted down anyway.

‍