On Meta, the creative is the targeting

A few years ago, a Meta Ads specialist earned their fee through audience research: interest stacks, lookalike percentages, layered exclusions. Most of that work is now done by the delivery system itself. I have spent seven years running growth and performance marketing, first inside B2B firms and now through my agency, and the shift I see in every account audit is the same: the people still fighting over audience settings are optimising the part of the machine that no longer listens to them.

That doesn't make the specialist less valuable. It moves the job. The targeting did not disappear; it moved into the ad. Your hook, your format, the person on screen and the first line of copy now tell Meta who to find. This piece explains how that works, and how to build a creative system around it.

Why the targeting moved into the creative

Start with the plumbing, because most advice skips it. In December 2024 Meta's engineering team published a detailed account of Andromeda, its machine-learning system for the retrieval stage of ad delivery. Retrieval is the first cut: from tens of millions of eligible ads, the system selects a few thousand relevant candidates for a given person, and later ranking models choose from that shortlist. Meta reported a 6% recall improvement and an 8% ads quality improvement on selected segments after deploying it, and it named the reason the system was needed: the volume of creative produced through Advantage+ automation was growing fast.

Then the campaign layer changed. In June 2025 Meta announced that sales and app campaigns would no longer be a choice between an Advantage+ shopping campaign and a manual one. Instead, a campaign is "Advantage+ on" when Advantage+ campaign budget, Advantage+ audience and Advantage+ placements are active, and the old shopping-campaign API was scheduled for full deprecation by early 2026. Automation is no longer a product you opt into. It is the default state of the account.

Put those together and the logic is plain. When the audience is broad and the budget, placements and audience expansion are automated, the system needs another input to decide who your ad is for. That input is the ad itself and the behaviour of whoever engages with it.

Your creative is no longer just the message. It is the brief you are giving the algorithm about who to find.

A tactical example: a skincare clinic runs one video in which a dermatologist explains treatment science, and a second in which a bride-to-be talks about her wedding date. Same offer, same broad audience. Those two ads will find different people, because the first earns attention from research-led buyers and the second from event-driven buyers. You did not set that targeting anywhere in Ads Manager. You shot it.

What creative diversification actually means

Meta's own Blueprint training, titled "Increase campaign performance with diversified creative", frames diversification along two axes: concept and motivators, and format. That is the right lens, and it is where most accounts go wrong. They produce volume and call it diversity.

Here is my first contrarian point. More ads is not more diversification. Twenty UGC clips of different creators saying the same script over the same product shots look like one idea to a retrieval system, and they will mostly reach the same pocket of people. You have paid for twenty productions and bought one test.

Format is the second axis, and it is underused. A static testimonial card, a founder talking to camera, a product demonstration with captions and a carousel that walks through objections are four formats that tend to attract different viewing behaviour. If every ad in your account is a 30-second talking-head Reel, you have told the system to find people who watch 30-second talking-head Reels.

The mistake to avoid is treating diversification as a production quota. The thing most teams miss is that diversification starts in the brief, not the edit. If the brief contains one idea, no amount of editing will produce five.

Build an angle library around personas

Write for one person. A specific problem, described in the buyer's own words, finds that buyer. This is the part of the original craft that matters more now, not less: audience research did not die, it moved from the targeting panel into the creative brief.

I build an angle library before a single ad is shot. It is a living document, usually a simple sheet, that maps who buys, why they buy now, and what would make them believe you. Each row becomes a testable concept.

1. List three to five buyer personasDefine them by situation and motivation, not demographics. "Owner who just lost a key salesperson" is useful; "male, 35–44" is not.
2. Capture their languagePull exact phrases from sales calls, reviews, support tickets, WhatsApp enquiries and comments. The words customers use become your hooks.
3. Map the buying triggerFor each persona, write the moment that makes them act: a deadline, a failure, a comparison, an event. Triggers make the strongest openings.
4. Choose the proofDecide what evidence each persona needs: a demonstration, a credential, a before-and-after, a price comparison or a peer story.
5. Assign a formatMatch the concept to the format that carries it best, then make sure your live mix covers several formats.

In practice, the most productive hour in any engagement I run is not spent in Ads Manager. It is spent listening to recorded sales calls. Your media buyer should probably be closer to your sales team than to your analytics dashboard, because the sales team hears the objections your ads need to answer.

AI tools can help here, for drafting hook variations or summarising call transcripts, but they work best on top of the library, not in place of it. I wrote more about where that line sits in AI in marketing workflows.

A testing system that learns every week

Paid media rewards the teams that learn fastest. The goal of a testing system is not to find one winning ad. It is to find out, repeatedly, which reasons to buy move your market, and to feed that learning back into the next brief.

Keep account structure simple. Fewer campaigns with more budget give the system enough conversion volume to learn. Splitting a modest budget across many ad sets, each with its own audience, starves every one of them and recreates the manual targeting that automation was built to replace.

My weekly rhythm has four steps, and it has barely changed in years:

1. Review by conceptLook at which concepts produced qualified results, not just cheap ones. Group the ads by the persona and angle tags in their names.
2. Retire what is clearly not workingPause concepts that have spent enough to judge and still show no sign of qualified response. Be decisive; dead ads dilute budget.
3. Launch two or three new anglesPull the next concepts from the angle library, informed by what last week taught you. New concepts, not new colours.
4. Write down the learningOne or two sentences per test in a shared log. The log is what makes the next test start smarter, and it survives staff changes.

The mistake to avoid is editing live ads every day. Constant changes reset learning and make results impossible to read. Decide on a review day, act on that day, and leave the system alone in between. This is the same discipline I describe in learn, build, test, repeat and, at a business level, in marketing systems over campaigns.

Signals decide who the creative finds

Creative tells the system who might care. Conversion signals tell it who actually bought. If the second half is broken, the system optimises your excellent creative toward the wrong people, quickly and at scale.

Meta's Conversions API documentation recommends running the API alongside the Meta Pixel, sending the same events through both, because the server connection can recover website events the Pixel misses through network or page-load problems. It asks you to deduplicate those events by sharing the same event name and an event ID, to send customer information parameters such as email and phone number to improve matching, and to send events as close to real time as possible. You can then check the Event Match Quality score, rated out of 10, in Events Manager.

Judge on business outcomes. Optimise toward the event closest to revenue that still gives the system enough volume to learn. For an e-commerce brand that is usually purchase. For a lead-generation business it is often a qualified lead, sent back from your CRM, rather than the raw form submission.

Reading results and managing fatigue

My second contrarian observation: the most dangerous ad in your account is often the cheapest one. A concept with a low cost per result can be attracting the wrong buyer, and because the system pushes budget toward whatever performs on the metric you chose, a cheap but poor-quality ad can quietly take over spend. Always check downstream: purchase value, lead-to-sale rate, refund rate, repeat purchase.

Read results by concept rather than by individual ad. A single ad's numbers are noisy; a concept's numbers across several executions tell you whether the underlying reason to buy is working. That is why the naming convention from earlier matters.

Fatigue is the other half of reading results. Signs worth watching include rising frequency alongside falling click-through rate, a climbing cost per result on a concept that used to be stable, and comments that repeat the same objection. The operational reality is that fatigue usually arrives at the concept level, not the ad level: swapping the creator or the background buys a little time, but the audience has already heard the idea.

The fix is not to panic-launch twenty new ads. It is to go back to the angle library, pick the next persona or motivator, and give the system a new idea to work with. Variation extends a winner; diversification replaces it.

Where to start this week

If you do one thing, audit your live ads and tag each one with the concept it represents. Most teams discover that a large active library collapses into two or three ideas. That single exercise tells you how diversified your account really is.

Then fix the signal: confirm the Conversions API is live, events are deduplicated, and you are optimising for the event closest to revenue. Finally, build a short angle library and commit to launching two or three new concepts on the same day every week. Within a month you will have a log of learnings that no competitor can copy, because it came from your market.

The specialist's job did not disappear when Meta automated the targeting. It became strategy, research and creative direction, which is where it should have been all along.

Frequently asked questions

Does detailed targeting still matter on Meta ads?

It matters less than it used to. When Advantage+ audience is on, the interests and demographics you add work as suggestions that the system can go beyond if it finds better results elsewhere. Exclusions such as existing customers and location rules still matter for many businesses. For most accounts, though, the larger gains now come from creative that clearly signals who the ad is for, and from conversion signals that tell the system who actually bought.

What is Meta Andromeda?

Andromeda is Meta's machine-learning system for the retrieval stage of ad delivery, described by Meta's engineering team in December 2024. Retrieval is the first filter: it narrows tens of millions of candidate ads to a few thousand relevant ones for a given person before ranking models make the final choice. For advertisers, the practical takeaway is that the system needs varied, distinct creative to match different people well, which is why diversification of concepts matters.

How many creatives should I test on Meta each week?

There is no official number from Meta, and the right volume depends on budget. A useful rule from my own practice is two or three new concepts per week for a modest budget, each with one or two executions, rather than a large batch of small variations. What matters is that each concept is a different reason to buy or a different format, so every test teaches you something new about your market.

Do I need the Conversions API if I already have the Meta Pixel?

Meta recommends using both together. Its documentation explains that the Conversions API can share website events the Pixel may lose through network connectivity issues or page loading errors. Send the same events through both, deduplicate them with matching event names and event IDs, and include customer information parameters to improve matching. Then check your Event Match Quality score in Events Manager to see how well server events are being matched.

How do I know when a Meta ad has creative fatigue?

Common signs include frequency rising while click-through rate falls, cost per result climbing on a concept that used to be stable, and comments repeating the same objection or showing that people have seen the ad many times. Look at these signals at the concept level, not just the single ad. If several executions of one idea are all weakening, the audience has heard that idea, and a new angle will usually help more than a new edit.

What is the difference between creative variation and creative diversification?

Variation changes the execution of one idea: a new headline, a different background or another creator reading the same script. Diversification changes the idea itself: a different problem, motivator, buyer persona, type of proof or format. Variation is useful for extending the life of a proven concept. Diversification is what gives an automated delivery system distinct options to match to different people, and it is where most of your weekly testing effort should go.

Sources

1. Meta Andromeda: Supercharging Advantage+ automation with the next-gen personalized ads retrieval engine — Engineering at Meta, 2024
2. Advantage+ campaign experience for sales and app — Meta for Developers, 2025
3. Best Practices — Conversions API — Meta for Developers, accessed 2026
4. Increase campaign performance with diversified creative — Meta Blueprint, accessed 2026
5. Advantage+ Audience Best Practices Guide — Jon Loomer Digital, accessed 2026