AI ads are here — and they're invisible

A version of this article originally appeared in Quartz’s AI & Tech newsletter. Sign up here to get the latest AI & tech news, analysis and insights straight to your inbox.

Mark Zuckerberg laid out Meta $META's advertising future on a recent earnings call, and it sounds like a marketer's dream. Advertisers will simply provide a business objective and payment information, he said, and AI will figure out everything else, including generating personalized video and creative content tailored to individual users.

For as much information as tech companies already have on us, the AI tools being rolled out could know us as intimately as friends. And unlike today's algorithms that track clicks and purchases, these systems will understand your insecurities, your aspirations, and exactly what it takes to change your mind. That psychological profile then becomes a product — think Google $GOOGL AdWords meets your therapist's notes — sold to the highest bidder.

Google is already testing ads in its AI chatbot responses. OpenAI is staffing up a new advertising platform. Ticketmaster is running AI-generated Facebook ads featuring virtual families whose team allegiances shift based on who's viewing them. Eventually, those could be families that look like yours, matching your demographics and characteristics. 

Then comes the logical next step of ads that use your own photos, digitally inserting your actual family into branded content. (Don't believe me? Facebook pioneered something similar back in 2009, using members' profile photos in ads shown to their friends when they became fans of brand pages.)

The infrastructure for hyper-personalized advertising at scale is being built right now, and it's designed to be invisible.

That's the problem. Unlike traditional ads that are clearly marked as sponsored content, AI-embedded advertising could emphasize certain topics or use particular language while maintaining the illusion of neutral helpfulness. When your AI assistant is financially incentivized through advertising revenue, it might steer conversations toward revenue-generating topics without you even noticing.

The technology can already deliver different experiences to different people

The business case driving this transformation is straightforward. Management consultancy McKinsey reports that personalization can reduce customer acquisition costs by up to 50%, lift revenues by 5% to 15%, and increase marketing ROI by 10% to 30%. Companies with faster growth rates derive 40% more revenue from personalization compared to slower-growing competitors.

Multimodal AI systems that process text, images, audio, and video simultaneously are eliminating the old tradeoff between personalization and scale. For decades, companies could create highly tailored experiences for small groups or reach massive audiences with generic messaging — but not both.

The Las Vegas Sphere already uses audio systems that let people standing inches apart hear entirely different content. Johnnie Walker ran an experience in Edinburgh where visitors answered three questions that generated entirely unique bottle labels printed within minutes. These are early examples of what becomes possible when AI can analyze multiple data streams and personalize experiences across every dimension simultaneously.

That makes manipulation almost impossible to spot. When everyone sees the same billboard or TV commercial, you can at least discuss whether the messaging feels manipulative. When the content is individually tailored and invisible to everyone else, there's no way to compare notes or call out problematic tactics.

Companies are pushing ahead despite fundamental problems

It's already working. One retailer McKinsey talked to for its report said it used AI to target promotions and saw sales rise as much as 2% and margins improve up to 3%. What that actually means: The system figured out which shoppers were price-sensitive enough to need a discount and which would pay full price regardless. You're getting a deal because an algorithm calculated you're desperate enough to need one, while the person next to you pays more for the same item.

Companies are investing heavily to scale this despite significant technical limitations. For example, AI struggles with analyzing darker skin tones because training data skews toward lighter complexions. It also has issues processing languages beyond the major ones that dominate training data. And of course, running complex personalization in real-time burns through expensive computing resources. Building infrastructure for true customization requires rethinking entire production processes, which is why most personalized experiences remain one-off demonstrations rather than everyday reality.

But those are solvable engineering problems, and the profit potential is too big to ignore. The unsolvable problem is transparency. When AI personalization becomes sophisticated enough, you won't be able to tell whether a recommendation reflects your genuine interests or has been optimized to benefit whoever paid for influence.

There's no way to see why you're being shown what you're being shown. And when everyone experiences different content, hears different audio, and gets different advice, there's no shared baseline to reveal when the algorithm is steering you somewhere profitable rather than helpful.