Sponsored recommendations just got a friendly, conversational makeover. New research from Princeton and the University of Washington reveals that when your interests conflict with platform profits, AI shopping agents consistently push the pricier option. Testing 23 language models in scenarios like flight booking, researchers found 15 models recommended expensive sponsored choices over cheaper alternatives that better matched user preferences—even when the budget option was objectively superior.

The Truman Show Economy

AI agents speak like helpful friends while potentially steering you toward whoever pays them.

This isn’t some distant dystopia. ChatGPT’s shopping tools, Perplexity’s “Buy with Pro,” and Amazon’s cross-site agents already handle product research and purchases for millions of users. They speak in caring, helpful language—your “personal shopper,” your “AI concierge”—while potentially steering you toward whoever pays them.

Like Meryl in The Truman Show, they play the loving assistant while pivoting to pitch products you never needed. The incentives are baked into the system. Retail media already commands 65% of ad spending through sponsored search results, according to Bain & Company. As shopping moves to chat interfaces, those same pay-to-play economics follow.

OpenAI claims ChatGPT’s results remain unsponsored, but long-term monetization models stay conveniently vague.

The Algorithm’s Hidden Hand

Columbia Business School researchers warn that whoever controls product rankings controls what gets bought.

AI agents don’t shop like humans—they obsess over rankings and structured data feeds. When agents scan thousands of products in milliseconds, slight algorithmic tweaks become powerful steering mechanisms. You think you’re getting objective analysis, but you’re seeing curated results optimized for someone else’s bottom line.

UK government analysis warns of “loss of consumer agency” as agents gain autonomy over purchases. Without transparency requirements or fiduciary duties, your helpful shopping bot might be more salesperson than servant.

The convenience is real—these tools save time and potentially money. But as AI agents handle more purchasing decisions, the question isn’t whether they’ll manipulate your choices. It’s whether you’ll notice when they do.