On Meta Ads, creative does the targeting

The instinct carried over from search advertising is that targeting is where you exert control and creative is what you put in front of the audience you selected.

On Meta that has been inverted for some time. The creative is the targeting instrument. Who responds to an ad tells the delivery system who to show it to next, and it expands toward that signal far more effectively than an interest list assembled by hand.

This is an argument, and reasonable people disagree with parts of it. The strongest objection is addressed below.

Why audience settings lost influence

Two shifts, both well documented.

Signal loss changed what targeting could do. Platform-level privacy changes — most visibly Apple's App Tracking Transparency — reduced the third-party signal available for building and matching audiences. Meta's response was to lean harder on its own delivery modelling and on first-party signal.

Automated audience expansion became the default direction. Broad targeting with automated placement and expansion is where the platform has been steering advertisers for several years, and the tooling is built around it.

The practical consequence: an interest stack that would once have meaningfully narrowed delivery now more often just constrains the pool the system can optimise within. You have not improved relevance; you have removed options from a system that was going to find those people anyway.

What still legitimately narrows delivery: exclusions, geography, language, age where genuinely relevant, and custom audiences built from your own data. Those remain useful. Stacked interests, mostly, do not.

Because the creative carries the targeting, the work shifts from audience research to producing enough distinct concepts to test — which is how we structure

creative-led acquisition with real attribution behind it.

What the ad is actually doing

When creative is the targeting mechanism, each ad becomes a filter.

An ad that names a specific problem attracts people with that problem. An ad that leads with price attracts price-sensitive buyers. An ad that shows a product in a particular context attracts people who recognise that context.

The system then expands toward whoever responded. So the creative does not merely communicate to an audience — it selects one.

That reframes the work. Instead of asking "who should see this", you ask "what would make exactly the right person stop, and exactly the wrong person scroll past". The second question produces better ads and better delivery at the same time.

Concepts, not variations

The most common testing mistake is running many variations of one idea and concluding Meta does not work.

Variations produce small differences. Concepts produce large ones, and only large differences are worth spending a test budget to detect.

A practical structure: a small number of genuinely distinct concepts, each with enough budget to accumulate readable data. Fewer, more different, better funded — rather than many, similar, and starved.

Volume beats polish

Creative fatigues. The reachable audience sees an ad enough times that response declines, and no bid adjustment fixes that. The signature is rising frequency alongside falling click-through.

The implication is that a pipeline matters more than any individual winner. An account producing new concepts steadily outperforms one that found a good ad and rode it, because the second is on a decline it cannot bid its way out of.

This is where production polish gets over-weighted. A clearly argued ad filmed on a phone regularly beats a polished production that says nothing specific. Polish helps once the message works; it cannot rescue a message that does not — and polish is usually what makes a creative pipeline too slow to sustain.

The strongest counter-argument

Broad targeting is not always right, and treating it as universal is its own error.

Three cases where deliberate targeting still earns its place:

• Small budgets. Broad needs volume to find the pattern. Below a certain spend, a constrained pool concentrates the signal enough to learn something
• Genuinely niche products. Where the addressable audience is small and identifiable, telling the system directly is faster than letting it discover them
• Exclusions, always. Existing customers, recent converters, unsuitable geographies. These are not optional and broad does not handle them for you

So the honest position is not "targeting is dead". It is that creative now carries more of the weight than audience settings do, and most accounts have the effort split the wrong way round.

What bounds all of this

Creative testing only works if the conversion signal is trustworthy. The system optimises toward the conversions it receives, so:

• If conversions are double counted, it learns from inflated data
• If a form fill is the objective when a sale is the goal, it optimises for form fillers
• If consent handling excludes a share of conversions, it optimises on a subset

Under any of those, a genuinely good creative can look mediocre and a poor one can look excellent. The measurement bounds the ceiling on everything above.

That is not a Meta-specific point — it is the same argument as conversion tracking that actually works, and it is the reason tracking is stage one of every audit we run rather than an afterthought.

Where to put the effort

1. Verify the conversion signal. Everything downstream inherits it
2. Set exclusions properly, then go broad unless a case above applies
3. Build a concept pipeline — several genuinely different arguments, not variations
4. Fund each concept enough to produce a readable result
5. Watch frequency and click-through together as the fatigue signal
6. Replace concepts before they decline, not after

The uncomfortable implication for most accounts: the work moves from the campaign interface into producing ads. That is a harder operational problem than adjusting audience settings, and it is where the results now are.

Keep reading

This argument connects to a few other things worth reading before you make a decision.

Browse resources