How might Meta’s AI ads actually work? Good question.
by Deacon Webster
“…We’re going to get to a point where you’re a business, you come to us, you tell us what your objective is, you connect to your bank account, you don’t need any creative, you don’t need any targeting demographic, you don’t need any measurement, except to be able to read the results that we spit out.”
– Mark Zuckerberg
This statement from Meta Founder and CEO Mark Zuckerberg generated a lot of chatter. Many in the ad world see Meta AI as a harbinger of the coming AI apocalypse where agencies are discarded in favor of all powerful AI’s that create and optimize themselves automatically without human intervention. Others, me included, are unconvinced as to whether this feature would be able to overcome the numerous hurdles that stand between the concept and the reality of marketers’ needs.
Some of my big questions:
How will the ads optimize themselves? For views? Clicks? What about reach campaigns? In any instance where the sale happens outside the tagged Meta ecosystem, the AI will have no data to inform its decisions. Then what? (Incidentally this is also is current day problem with Meta ad optimizations, but you won’t hear it from them).
Will marketers even know what creative Meta is showing? When Meta makes a change based on data, who sees that change and when?
How much leeway will the platform have to make whatever ad it wants?
And, interestingly, if given a long leash, what kind of ads would Meta’s Ad AI make? On that front, we can make some guesses based on what we know about Meta “best practices” and the way it serves content:
A common meta “best practice” to show a logo/product shot in the first three seconds of a video because users typically only watch the first three seconds. Thanks to Les Binet, Grace Kite, and Tom Roach, we know that this is exactly the wrong advice because ads viewed for under five seconds do not make it into retained memory, so it’s as if your ad never ran – but let’s assume Meta doesn’t give a shit about that (they don’t). We know that Meta also LOVES a strong call to action, and it has never seen a question in the post copy that it didn’t like (you know, ’cause engagement).
So, every ad will feature some sort of opening logo/product shot (ideally both), persistent logo bug, some optimized ad stuff based on Meta’s stores of data, and a big loud call to action.
Which “optimized ad stuff”? That’s the secret sauce that Meta gets from all its proprietary data. Given what we know about the types of content Instagram’s “explore” recommendations typically serve up, it’s a safe bet that that within days of launching this feature, every ad on Meta will look like this:
It’s not hard to imagine that left to its own devices, Meta’s AI would quickly turn every advertiser into a purveyor of soft-core porn.
If we assume that at some point Meta would figure out how to temper the basest urges of the system, we should still expect a steep rise in ad homogeneity. AI models are great average-ers; they summarize. Feed into the system every ad that’s ever run on Facebook and then ask it to come up with one more; it isn’t going to endeavor to create something new and unique. That’s not how it works. Everyone’s going to have the same ad with minor alterations based on “brand assets”.
What about nuances between similar brands? Is Meta AI going to be equipped to parse the difference between the Lexus brand and the Infinity brand? Samsung and LG? Coke Zero and Diet Coke? Does that even matter in an AI optimized ad-scape?
What of the vast majority of people who choose NOT to click or comment on ads in meta? What of the 95% of people who aren’t currently in the market for whatever advertised product they’re served? How will meta optimize for audiences that generate no actionable data?
What happens to brands? We know that DR type ads have a negative effect on brand perception. What about AI generated DR ads with a neon magenta blinking “BUY NOW” button?
And of course, would any advertiser be so naive as to trust Meta’s self-reporting given its history of falsifying everything from audience data to video view times?
It’s easy to see why Mark Zuckerberg would love the idea of marketers connecting their bank accounts directly to Meta; even better when all they have to offer in return is a spreadsheet at the end of the month. It’s more difficult to determine which advertiser would accept this arrangement.
