Advertisers who let Meta generate and keep adjusting their ad creative may end up with better campaign numbers and less understanding of why, according to a post Brainlabs published today by Liz DeAngelis, the agency's Managing Director, Programmatic & Media. The post, titled "If Meta Keeps Optimizing Your Creative, What Are You Actually Learning?", uses Meta's Muse Image model as its example and argues that what brands learn from creative should travel to other channels and future briefs rather than stay inside one platform. DeAngelis does not tell clients to stay away from the tool. She would test it, she writes, but she would not hand it a whole campaign on day one.

The piece is a column, not a study. It contains no campaign data, no client results and no response from Meta. Its one factual claim about the product concerns timing: Muse Image, according to Brainlabs, "is still on its way to Advantage+ creative," Meta's set of AI tools that generate and alter ad assets inside its Advantage+ automation suite. Meta launched Muse Image on July 7, 2026, and days later said advertisers and agencies would get access through Advantage+ creative within weeks, as PPC Land reported on July 12. If the Brainlabs description is current, that integration has taken longer than the weeks Meta indicated: 125 days separate the launch from today, by PPC Land's count.

In Short

A senior Brainlabs executive wrote today that when Meta's AI makes ads and keeps changing them, brands may see results improve without learning which image, message or format caused it. That matters for any advertiser that uses what it learns on Facebook and Instagram to brief its creative team or to plan TikTok and YouTube campaigns, because those lessons may stay locked inside Meta. Her answer is not to refuse the tools but to test them on a small, defined slice of activity with a clear comparison, while the arrival of Muse Image in Meta's ad tools is, by her account, still pending.

A landscaper in the garden

DeAngelis opens with an image. "Meta is bringing in a landscaper to their walled garden," she writes, casting Meta as a walled garden - a company that sells ads on its own properties, using its own logged-in data and its own measurement - and Muse Image as the hired help.

She does not dispute the appeal. "More creative variation, faster production, less resource needed to keep campaigns fed with new assets," she writes of the tool's promise. "For brands that struggle to produce enough creative, I think it's absolutely worth a test." The objection comes in the next breath: "the more creative decisions we hand over to platforms, the less we may understand about what's actually working."

Why testing produced knowledge

Her case rests on what creative testing has traditionally been for. A team changes one variable, watches the result and carries the finding into the next piece of work. She lists the candidates: "the hook, the product shot, the format, the message or the CTA." Whatever the variable, something from one campaign feeds the next.

Continuous change breaks that loop. "If Meta is generating variations and tweaking creative throughout a campaign, without an approval process or detailed reporting on exactly what changed and why, we might know that performance improved without knowing what actually improved it," DeAngelis writes. When something is always changing, the ability to isolate "the thing that worked in the first place" shrinks. Eventually, she argues, creative learnings start to look "less like actual learnings and more like 'Meta found something that worked.'"

The mechanism is not new in kind. In dynamic creative optimization, ads are assembled at the moment of serving from a template and interchangeable parts, and delivery shifts toward the combinations that perform. What changes with generative models is that the parts themselves can now be produced by the platform, in volume. Meta's developer page for Muse Image, which the Brainlabs post links to, lists ad variant generation, catalog imagery and per-user personalization among its uses and prices the model at $0.01 per image.

Lessons that should leave the platform

The stakes, in her telling, go beyond any single Meta campaign. Creative learnings "should help brands get smarter about their audiences," she writes, and should shape work on TikTok and YouTube as well as "what goes into the next brief" and what a creative team produces. The garden metaphor returns: "If all of that learning stays inside Meta's walled garden, we might be making the garden nicer, but I'm less convinced we're becoming better gardeners."

Where she draws the line

DeAngelis is explicit that generative creative is not her target. "To be clear, I think AI-generated creative has a lot of value for production and optimization," she writes. "Where I draw the line is assuming that means the process needs to happen inside the media platform."

Her preferred arrangement keeps generation at arm's length from media buying. "I'd much rather brands have their own relationships with AI and creative partners, decide their own comfort levels, and manage that process outside the platform," she writes. That, in her view, gives brands more control over what gets produced and a better chance of understanding what they are learning. Every brand will have a different appetite for AI-made creative, and each should be able to decide for itself "rather than being pushed toward whatever a platform decides is the best way to optimize."

Agency-built tools of this kind already exist. In October, Meta published a case study on Nest Commerce, a UK agency that runs its own Nest AI Creative tool, and credited Nest's AI ads with 20% higher conversion rates than non-AI creative - a figure PPC Land noted came without a stated measurement method or control group. Brainlabs itself lists paid social and performance creative among its services, so the model DeAngelis favours is also one in which agencies such as hers keep a role in production. The post does not address that point.

One company, four jobs

The broader question, for DeAngelis, is concentration. "Meta already has an enormous amount of performance data. It delivers the media, controls the optimization and reports back on the results," she writes. As Meta moves into generating and adapting the creative, another part of the process "moves behind the same walls." Her summary: "Creative generation, media delivery, optimization and measurement can increasingly happen within one ecosystem."

She grants the case for it. "Meta has access to a scale of performance data that an individual advertiser doesn't, and using that data to make faster creative decisions could create efficiencies that would be difficult to replicate manually." But advertisers, she writes, should understand the trade. "The more Meta learns on your behalf, the more important it becomes to ask how much of that learning actually makes its way back to you."

The scale she describes is visible in Meta's own numbers. Nearly 10 million advertisers used at least one of Meta's generative AI creative tools in the third quarter of 2026, up from more than 8 million in the first, as PPC Land reported from Meta's third-quarter earnings call. In late October, Meta said those tools had helped advertisers create hundreds of millions of new ads, enlarging the pool that its Andromeda retrieval system sorts before ranking, and claimed the system had lifted Facebook conversions 2.4% since 2025. The measurement side has drawn the same kind of question before: PPC Land asked in April who benefits when a platform builds the marketing mix model that helps allocate budgets to its own channels.

Control has also been tested in practice. In June, REI said Meta had auto-enrolled it in an AI personalization tool that altered a vendor-supplied bicycle image, leaving it with an extra set of handlebars and a missing crank arm. In October, PPC Land reported that Meta's revised Commercial Terms included a section that would make advertisers solely responsible for its AI ad edits, including edits from features Meta switches on itself. Meta then moved the start of those terms to November 20, and on November 5 paused the updates altogether, leaving its existing terms in effect, according to Nieman Lab.

Efficiency against growth

The obvious rebuttal is performance: if handing over control makes campaigns more efficient, does it matter that no one understands each decision? "Maybe," DeAngelis concedes. "But I think we have a challenge proving that these tools actually improve performance over time."

She expects platform evidence to arrive regardless. "I'm sure there will be some vanity metric or source to say it's working (as platforms tend to do)," she writes, before posing what she calls the harder question: "where is the brand that has genuinely driven incremental year-on-year growth because it handed more of its creative decision-making over to the platform?"

That bar, she notes, sits well above making one campaign more efficient. "My concern is that we could end up gaining efficiency at the expense of sustainable performance and growth," she writes. "A system can get very good at optimizing toward the signals it can see, but if the advertiser comes out of that process with less understanding of its audience and fewer learnings it can apply elsewhere, I'm not sure we should automatically call that progress."

Meta has not lacked for figures. It has reported a 22% return on ad spend increase for Advantage+ creative users and a 7% conversion increase from image generation, numbers that featured in PPC Land's coverage of advertiser skepticism toward its AI automation in late 2025. In October 2026 it said its GEM ads model had lifted conversions 18.9% across its apps, without giving a baseline, period or method. At Cannes in June, it put downstream revenue at $4.13 per dollar of Meta ad spend. None of these measures year-on-year brand growth, the test DeAngelis sets. Incrementality testing, which compares outcomes against what would have happened without the ads, can produce a much smaller number: one such test, cited in PPC Land's October 2025 analysis of Advantage+, credited the product with 17% of the conversions the platform had reported.

How she would test Muse Image

The post stops short of rejection. "This is also why I don't think the answer is to avoid tools like Muse Image altogether," DeAngelis writes. "If a client asked me whether they should use it, I'd say yes, test it, particularly if creative volume and variation are already a challenge." She adds: "Everything is worth a test when there's a clear reason for testing it. I just wouldn't hand over the whole campaign on day one."

Her method starts with three steps. "I'd start with a specific subset of activity, establish what we're trying to learn before the test begins, and make sure there's something meaningful to compare it against," she writes. A fourth condition is about attitude rather than procedure: going in prepared for the test to fail. "Testing new technology means being comfortable with some of that investment struggling or failing; otherwise, it isn't much of a test."

What Meta already offers

The post's complaint centres on two gaps: the absence of "an approval process" and of "detailed reporting on exactly what changed and why." Meta has shipped partial answers to both, though the post does not mention them.

On July 11, 2025, Meta added a Creative breakdown to Ads Manager that reports results for each element in flexible format ads, which can hold up to 10 images and videos, and lets advertisers compare AI-generated image variations against the original. At Cannes in June 2026, Meta said a creative approval workflow was in testing. Muse Image itself launched with presets, the ability to reference a user's own uploaded photos and a markup tool for edits. And when Meta extended its Muse agent to small businesses on September 29, 2026, it said nothing would publish, send or spend without the owner's approval, as PPC Land reported.

Whether any of these would cover the variations Muse Image produces inside Advantage+ creative, and in how much detail, the post does not say. It also gives no detail on how Brainlabs expects the integration to work.

Why this matters for marketers

The learning DeAngelis wants to protect was thin before generative tools arrived. An EMARKETER report produced with Perion, covered by PPC Land in April, found that 89.2% of 111 US agency and marketing professionals considered creative important to campaign performance, but only 3.6% said creative performance was well understood and actively optimized, and 72.1% waited for results to decline before changing creative. On those numbers, many advertisers had little systematic creative learning to lose.

Where cross-platform learning does happen, it tends to sit outside any one platform. A whitepaper from Kroger, Vidmob and MMA Global, covered by PPC Land on October 4, analysed 1,934 ads across Meta and DV360 with an external model and found that about half of Kroger's investment had gone behind weaker-scoring creative. That is the kind of portable finding DeAngelis argues brands should keep.

The tension will not end with one product. "Muse Image is one product, but the choice it represents is going to keep coming up," she writes. Platforms will improve at generating creative, optimizing it and deciding on advertisers' behalf, and in many cases letting them do so will make sense. "The question for advertisers is how much control they're willing to relinquish in exchange for that efficiency, and whether they're still learning enough to make better decisions themselves."

Her conclusion is short. "AI can absolutely help us make creative faster and at greater scale. I'm here for that. I'm just not ready to outsource the learning with it."

Timeline

Summary

Who: Liz DeAngelis, Managing Director, Programmatic & Media at Brainlabs, writing about Meta and the advertisers who use its AI creative tools on Facebook and Instagram.

What: An opinion post arguing that when Meta generates and keeps adjusting ad creative, advertisers may see performance improve without knowing what caused it, and that the learnings may stay inside Meta. DeAngelis says AI-generated creative has value but should be managed outside the media platform, questions whether any brand has achieved year-on-year growth by handing creative decisions to the platform, and says she would test Muse Image on a defined subset of activity rather than a whole campaign.

When: Brainlabs published the post today. Muse Image launched on July 7, 2026, and Meta said in July that advertisers would get it through Advantage+ creative within weeks; according to Brainlabs, it is still on its way.

Where: Meta's advertising platforms, specifically Advantage+ creative, with the argument extending to how brands carry creative lessons to TikTok, YouTube and future briefs.

Why: Meta now delivers the media, runs the optimization, reports the results and increasingly produces the creative. With nearly 10 million advertisers already using its generative tools, the question of how much of what Meta learns comes back to advertisers affects how brands brief, test and plan across channels.