Advertisers are disabling Meta's AI creative tools on the grounds the "aggressive" rollout prioritises the platform's own adoption goals over brand control, according to agency executives.
Susan Harangi, paid media supervisor at Edge Marketing, said the pace of Meta's AI rollout made it difficult to see the benefits.
"From an agency perspective, the rapid push towards these tools, with their lack of consistency and the additional QA required, feels more aligned with Meta's broader AI adoption goals," said Harangi.
A key frustration was the opt-out rather than opt-in model.
Harangi said a number of Meta's AI creative enhancements were enabled by default, and that ads built before certain features launched could later have enhancements applied retroactively.
"They are automatically applied unless we opt out,” she said.
“We have also found that ads that were built and launched before these AI features became available can later have a number of enhancements enabled, requiring us to actively review and manage them.
"As a best practice, we disable most features to maintain control over the creative and ensure brand guidelines are being met."
Compared to other platforms, Harangi said Meta's approach was both more aggressive and less predictable.
"Meta's AI toolset is more aggressive and frequently changing than what we're seeing on other platforms,” she said.
“While other platforms also use AI to automate and optimise campaigns, Meta's enhancements are less consistent across accounts and require more manual oversight.”
That unpredictability did not mean the tools were without value, but Harangi said realising that value required tight conditions.
Edge Marketing ran a controlled test for an e-commerce client in which Meta's AI-generated background was applied to one ad variation while keeping the product image and text overlay unchanged.
The AI variation received twice the spend and generated five purchases against zero for the original over a one-month period.
"While we wouldn't draw a broad conclusion from this single test, it was a good example of where selectively using AI can deliver a positive result without compromising creative quality or brand consistency,” Harangi said.
But not all agencies had taken such a controlled approach, with Harangi explaining some agencies were already scaling back their use of Meta's AI.
“From what I’m seeing, agencies scaling back Meta’s AI tools are doing so after issues with uncontrolled enhancements have impacted their client relationships.”
Ori Gold, co-founder and CEO at Bench Media, said the problem was not the tools, but the way agencies were deploying them.
"The mistake is treating all of Meta's AI as one capability," said Gold.
"Meta is very good at deciding which approved ad to show to which person.
“It is not yet consistently reliable enough to decide what the brand should say or how its ads should look. We are increasing our use of the first and maintaining tight human controls around the second."
That distinction shaped how Bench Media approached the platform.
On the optimisation side, bidding, budget allocation, placements, audience expansion and creative selection, Gold said Meta's tools performed well, particularly when fed sufficient conversion data and a strong range of approved assets.
The generative creative layer was a different story.
"The generative creative layer is less mature. Product accuracy, brand tone, disclaimers and local context cannot be treated as minor details,” he said.
“A variation may produce a stronger click-through rate while still being wrong for the brand or misleading to the customer."
Gold cautioned against taking Meta's own reporting at face value.
"We also need to be careful about allowing a platform to both deliver and grade its own automation. Platform-reported ROAS is useful evidence, but it is not the whole answer,” Gold said.
“We still assess results against broader business outcomes and incrementality."
When Meta's AI automation produces unacceptable output, Gold said Bench Media had already worked out a clear correction process, pause the ad, disable the enhancement, restore the approved asset and check live delivery again after republishing.
"The issues are not always dramatic,” said Gold.
“More common examples include a crop obscuring a disclaimer or logo, a text variation changing the meaning of an offer, or an image adaptation creating an off-brand representation of the product.
“It does not happen on every campaign, but it happens often enough that checking these variations is now part of our standard QA process."
Gold said the tools were not saving time so much as redistributing it.
"The irony is that a tool designed to reduce production time can simply move that time into review,” he said.
“If platforms continue producing more variations at greater scale, agencies and clients will eventually need to recognise that governance work explicitly within scope."
David Halter, chief practice officer at Dentsu, experienced the same tension between the Meta tools and time saving.
"It can dramatically accelerate variation, adaptation, testing and optimisation," said Halter.
"But if you can't trust the output, you simply move the time you saved in production into QA. Good automation removes work. Bad automation just relocates it."
For Halter, the fix came down to three things Meta had not yet delivered: transparency about what was being changed, the ability to protect non-negotiable brand assets, and confidence that what was approved was what actually appeared in market.
"Clients and marketers need to know what is being changed, be able to protect non-negotiable brand and product assets, and have confidence that what they approve is what ultimately appears in market," he said.
The stakes of getting that wrong were not trivial. Halter said brands and advertisers had forgotten responsibility lay with them, regardless of a machine decision.
"Two handlebars or an extra limb might be funny in an industry article — an inaccurate product, price or brand claim appearing in front of a customer is a very different matter,” he said.
“Ultimately the brand and agency remain accountable for what goes into market."
Beyond the immediate operational issues, Halter said the more consequential question was what happened when every agency had access to broadly the same AI capabilities.
"AI can give you a hundred versions of an idea. Someone still has to have the idea. As execution becomes easier, differentiation becomes harder. That's why I think creativity matters more in an AI world, not less," he said.
"I don't want us to automate the joy out of the job.
“The debate, the challenge, the tough conversation, the moment you put an idea in front of a client and see their reaction — that messy process is often where the magic happens.
“AI should remove the drudgery and make great thinking work harder. It shouldn't remove the thinking, or the joy of making something great."
Have something to say on this? Share your views in the comments section below. Or if you have a news story or tip-off, drop us a line at adnews@yaffa.com.au
Sign up to the AdNews newsletter, like us on Facebook or follow us on Twitter for breaking stories and campaigns throughout the day.