What Meta’s Ad Automation Needs From Your Creative

In June 2025, the Wall Street Journal reported that Meta aimed to let brands fully automate the creation and targeting of their ads with AI by the end of 2026: supply a product image and a budget, and Meta’s systems generate the images, video and copy, pick the audience and suggest how to spend. Mark Zuckerberg had been describing the same destination for months, a business stating its objective and the platform doing the rest.

Whatever arrives on that timeline, the direction is already built into how Meta delivers ads. Targeting has mostly moved inside the machine. What hasn’t moved is the question of what the machine is given to work with. Meta’s automation decides who sees which ad largely by reading the ads themselves, so the creative you supply has become the main input you control. The practical question is how much of it the system needs, and what kind.

Why the ads have become the targeting

Meta’s delivery works in stages. Before any auction, a retrieval system chooses a few thousand candidate ads for a given person out of an enormous pool, and ranking models then decide which wins. Meta rebuilt that retrieval stage with Andromeda, covered in Andromeda and the end of audience targeting as you know it, and has since put larger AI models behind ranking, including a foundation model it calls GEM, as this operational review of Meta’s engineering disclosures summarizes.

The consequence for an advertiser is simple to state. When the system matches ads to people by what the ads contain, each genuinely different ad is a different way to reach a different kind of person. An account with one idea expressed in twenty ways is, as far as matching goes, an account with one idea. An account with eight distinct ideas has eight routes to eight kinds of buyer.

Meta’s own recommendation to advertisers after the Andromeda rollout was creative diversification. Jon Loomer’s reading of that advice, in his breakdown of creative diversification, is the useful correction: making more ads is not the point. More ads only help if they represent a more diverse set of creative.

Variation is not diversity

This is the distinction most creative plans get wrong. Variation changes the surface of an idea: a different background, color, headline wording, crop or button. Diversity changes the idea: a different person the ad speaks to, a different problem it opens on, a different reason to buy, a different format or visual language.

Meta’s own generative tools are very good at variation. They generate backgrounds, adjust text, resize for placements and produce versions of an ad you already made. That is exactly the work the machine can do without you. What it can’t do is know that your product has a second audience you haven’t addressed, that a customer objection deserves its own ad, or that a testimonial would land harder than a demo for a certain kind of buyer. Those are concepts, and concepts are where your team’s time belongs.

A practical test: two ads are different concepts if a viewer would describe them differently to a friend. “An ad about saving time on payroll” and “an ad where an accountant complains about spreadsheets” are two concepts. The same ad in blue and green is one.

Be wary of the precise numbers circulating about this. Claims of a creative similarity percentage above which Meta merges your ads, or of an exact number of ads that lifts return by a fixed amount, mostly trace to vendors rather than to Meta. The same review notes there is no creative similarity score in Ads Manager, whatever confident guides suggest. The direction is well supported. The decimal places are not.

How much creative the machine actually needs

The honest answer depends on budget, because every concept needs enough spend to be shown, learned from and judged. Too few concepts and the system has little to match with. Too many for the budget and each one gets too little delivery to prove anything, which drags the account back into the fragmentation problems covered in how to lower ad costs.

So plan from the budget down rather than from a target count up. Work out how many distinct concepts your spend can realistically feed at once, keep that many live, and replace them as they fade rather than piling more on top. For each concept, supply a few formats, such as a short video, a static image and a motion piece, because formats reach people in different placements and moods, and let Meta’s tools handle the surface variants within each.

That shifts where production effort goes. Fewer rounds of small tweaks to one hero ad, more investment in a steady flow of new angles. Motion and short video are often the hardest formats for in-house teams to make at that pace, which is where motion graphics and animation built for ad testing earns a place in the plan.

Judge concepts on what they cause

When the machine chooses who sees what, it also shapes which ads look like winners. An ad the system shows mostly to people who were already close to buying will report strong results whether or not it persuaded anyone. Judge concepts on outcomes you can verify outside the platform where possible, and test the big calls with holdouts, the approach set out in what to measure instead of last click on Meta.

Keep a record of every concept, who it was for and how it performed. Over time that record tells you which kinds of ideas work for your brand, which is knowledge the platform will not hand back to you. Running the creative pipeline and the media together, so each informs the other, is the job of performance marketing that treats creative as the main lever.

Meta is automating the parts of advertising that can be computed: who, where, when and which variant. What it can’t compute is a new reason for someone to care. The advertisers who do well under automation will be the ones who keep supplying those.

Frequently Asked Questions

How many ads should I run in a Meta campaign?

There is no fixed number that suits every account. Run as many genuinely distinct concepts as your budget can give meaningful delivery to, each in a few formats, and replace them as they fatigue. Too few concepts leave the system little to match with. Too many for the budget spread spend so thin that no ad gathers enough data to be judged or optimized.

What is creative diversification on Meta?

It means giving Meta’s delivery system ads that differ in their underlying idea, not only in surface details. Different audiences addressed, different problems or reasons to buy, different formats and visual styles. Meta recommended it after rebuilding ad retrieval with Andromeda, because its system matches ads to people partly by what the ads contain, so distinct concepts reach distinct kinds of buyers.

Can Meta’s AI tools create my ads for me?

Partly. Meta’s generative tools can produce variations of ads you supply, such as new backgrounds, text versions and placement formats, and Meta has stated an ambition to generate complete ads from minimal inputs. The tools are strongest at variation. The concepts, meaning who the ad is for and why they should care, still come from people who understand the product and its buyers.

Does Meta merge similar ads together?

Meta hasn’t published a rule for merging similar ads. Its guidance to diversify creative implies that ads which are very alike add little extra reach, which is why variety of concept matters more than volume. The specific similarity scores and percentage thresholds quoted in many guides are not published by Meta, and there is no similarity score in Ads Manager. Plan around the principle rather than those numbers.