Facebook ads targeting: your creative is now your real targeting

We still spend way too much time tweaking audiences in Ads Manager. Age, gender, interests, lookalikes. And while we're busy doing that, Meta honestly couldn't care less. It looks at your creative, breaks it down, and that's what decides who your ad gets shown to. Not your settings.

I'm going to walk you through how Facebook ads targeting actually works today, why interests and lookalikes are dying, and most importantly a tactic we're testing agency-side to go after much cheaper CPMs on audiences nobody is fighting over.

How Facebook ads targeting actually works today

Quick refresher on what you're allowed to set in Ads Manager. There are three big families. Core audiences, built from Meta's native data, like geo, demo, interests, behaviours. Custom audiences, built from your own data, your site visitors, your email list, people who've already engaged with you. And lookalike audiences, the famous ones, where Meta hunts for people who resemble your buyers.

That's the theory. Now the reality.

When you launch a broad campaign, Meta runs a multi-step process. No magic in there, just well-oiled mechanics.

Step one, it analyses your creative. The audio, the transcripts, the visual shots, how people watch it in those very first seconds. It turns all of that into a vector, a kind of mathematical signature, and drops it into a very tight little group of buyers. A node, to use the term.

Step two, if your creative performs in that first group, say moms with toddlers who stop on your video, Meta tests it on a neighbouring group. Moms with slightly older kids, for example. It builds bridges between audiences based on the purchase signals they share.

Step three, and this is the point that changes everything in practice, the algo doesn't give a damn whether the person is 22 in Lille or 45 in Bordeaux. If both stop on the same type of content, they're in the same node. Period. Your demographic boxes, it skims right over them.

Step four, interests update in real time based on what people are doing in their session. Someone who spends ten minutes on "how to keep your kids busy" content today, their profile shifts within the day. And your ad finds a cheap way in, no matter what their long-term tags say.

That's why I tell our clients to stop micro-managing the settings. Your creative decides who your ad gets shown to. It's become your primary targeting variable. If the whole audience-control topic interests you, we wrote a dedicated piece on letting go of control on Meta.

Why interests and lookalikes are dying

It used to be you'd set an audience in Ads Manager. Tick "yoga interest," "lookalike 1% of purchasers," and let it run. That was your targeting.

That doesn't hold up anymore. Meta has so many real-time creative signals that an audience frozen in a dropdown menu quickly becomes stale info. Since consent got tighter, the pixel sends back far less data before acceptance, so the algo leans even harder on what it understands from your creative and session behaviours. That's why we're all moving to server-side tracking now. Without it, Meta optimises almost blind.

I'm not saying lookalikes died overnight. On some accounts they still get the job done. But the trend is clear, and I'd rather be straight with you than sell you ten-layer custom audiences that barely matter anymore. Today everything comes down to creative and angles. That's actually the core of what we do with our paid media offer for fashion and lifestyle brands.

And what about retargeting in all this? Re-engaging people who visited your site without buying still makes sense. But we keep it as tight as possible. We mainly treat Meta as an acquisition platform. With retargeting, you end up going in circles on a small audience, you burn it out, and you pay to reach people who already had you in mind. The goal is to go find fresh traffic cheap. And that's where you stop torching budget for nothing.

Node hopping, the thing nobody is exploiting

Truth is you don't have to fight inside your main node.

Your main node is shared with every direct competitor bidding in the same spot. High CPMs, audience overlap everywhere, everyone slugging it out over the same people. End result, you pay a premium to reach a person ten other brands also want.

Node hopping means building a creative that jumps into a neighbouring node. An adjacent interest group, not saturated, that holds exactly your buyer but at a fraction of the bidding cost.

It works because in these neural networks, every user, every brand, every topic exists as a vector in a space with tons of dimensions. Two topics don't need to be in the same product category to resemble each other. What matters is that the same group of people interacts with two creatives, even visually different ones. When that happens, the algo builds a bridge between the two nodes.

A concrete example with a dead-simple product

Take a cocktail shaker. Basic product, the kind of thing you don't even notice on the shelf.

Its main node revolves around the home bar and mixology. Expensive CPMs, competitors everywhere. You'll struggle and pay full price.

Now look at the neighbouring nodes you can go after.

First neighbouring node, premium protein shakers. You run a creative with a hook like "why I ditched plastic shakers for my whey and switched to a double-walled stainless steel shaker." Same product, different angle, different audience, lower CPM.

Second neighbouring node, specialty coffee and home espresso. You borrow the visual codes of that content, the ultra-polished lighting, the ASMR close-ups on metal, the measurements precise to the gram. And you apply them to making a cocktail. The same buyer, a precision freak, lives in both worlds.

Third neighbouring node, everyday gear and well-designed objects (the famous "everyday carry"). People who love durable stainless steel objects that last ten years. Your shaker fits right in there.

Three nodes, three completely different creative angles, one single product. And each time you go after the same type of buyer instead of fighting in the saturated arena.

How to test this without winging it

Still have to do it cleanly, otherwise you burn budget for nothing.

The method we use: we set up a testing campaign. One angle, one creative, an equal budget on each neighbouring node we want to explore. We let it run, we watch which angle delivers the best CPM and the best engagement rate. On a client account in home gear, the specialty coffee angle gave us a CPM 30% below the mixology node. So we move to the next iteration. We keep the winning angle and test several creative variants on it.

A word on arbitrage, because this is where most people trip up. When you run several angles in parallel, you have to decide which one you cut and which one you scale. We look at two things. CPM, to know whether the node is expensive or not, and the engagement rate on the first seconds, to know whether the creative really speaks to that audience. An angle with a low CPM that nobody watches is useless. An angle that engages hard but at an insane CPM doesn't hold either. You keep the one that checks both boxes, you cut the rest, you put the budget there.

Another important point, change one variable at a time. If you change the angle, the photo, the format and the hook all at once, you'll never know what worked. You run campaigns and learn nothing. You change one variant. Period.

And before you even launch, take two seconds on the numbers. Your ad budget should be built from your margin, not your revenue. If you go chasing cheaper CPMs but don't know how much you can afford to pay for an acquisition, you're flying blind. We cover it in detail in build your ad budget from your margin.

What I want you to take away

Stop spending your evenings fiddling with audiences in Ads Manager. That's not where the game is anymore.

Targeting happens in the creative now, the dropdown menu takes a back seat. Your angle and your hook decide which node your ad lands in. So you stop asking yourself which audience to tick. The real question is which angle to produce to catch your buyer where nobody else is looking.

Start small. Take your product, list three neighbouring nodes, write a specific hook for each, and test at equal budget. Then look at where your cheapest conversions fall. On our accounts, it's rarely the node we bet on at the start.