Nearly nine in ten AI shopping answers cite at least one commercially disclosed page, yet almost none of those disclosures survive into the answer itself. That figure — 86% of answers citing a commercial source, with roughly a quarter of cited pages carrying a disclosure — comes from a GetCited study of ChatGPT, Google's AI Overviews and Perplexity. It frames the real question for practitioners: not whether AI answers need "sponsored" tags, but whether the agents themselves are loyal to the user.

What do consumers actually tolerate?

People accept ads as the price of free services. What they punish is deception. A study of native ads on news sites found fewer than 10% of consumers identified them as advertising, and a Contently survey found two-thirds of readers felt deceived once they discovered an article was brand-sponsored. That history built the FTC's disclosure regime.

Shopping behavior has already moved. According to an impact.com survey, 47% of U.S. shoppers who use AI started their last product search with an AI tool. Meta says shopping is the most common use of Muse, its new personal AI agent. Notably, 24% of shoppers in the impact.com survey said they use AI specifically to avoid ads and sponsored results.

Would labels in AI answers even work?

A label could confuse users into thinking the platform was paid rather than the publisher — a real conflict for ChatGPT, whose ads appear separately and carry their own sponsored labels. OpenAI's current answer is the citation link: since launching ChatGPT search, it has included source links so users can verify answers themselves.

But users rarely check. Pew found people clicked a link inside a Google AI summary on just 1% of visits. On Gen Z, impact.com found 11% never click through while 21% always do. Asking an AI for air fryer recommendations is closer to hiring a researcher than reading a billboard — and you would not ask a researcher to stamp every brochure they touched.

Why affiliate content complicates the label debate

Any buying query sends AI toward product reviews, and those pages typically monetize through affiliate fees. FTC endorsement guidelines require sites to disclose affiliate relationships, and outlets like Wirecutter do so on every page. But if those disclosures carried into AI answers, they would appear in almost every shopping answer — a UI nightmare that would dilute the label's meaning. One possible compromise: a user-side "flag paid sources" toggle.

What about the platform's own kickbacks?

This is the sharper risk. A University of Michigan study found people rarely noticed when a chatbot wove ads into responses, and many preferred those answers; even with labels, about half the participants missed them.

OpenAI's ad principles state plainly: "ads do not influence the answers ChatGPT gives you." Its ads run only in the free and Go tiers, always separate from the answer. Perplexity made similar promises, then exited advertising earlier this year. One executive explained: "A user needs to believe this is the best possible answer."

The conversion data shows why the wall matters. Early research from the Frankfurt School of Finance & Management found shoppers who clicked ChatGPT ads bought at 0.37%, versus 2.21% for those arriving through regular answers. The moment answers look bought, they will convert like ads — which pushes brands to prioritize presence in answers over advertising, including among paying subscribers who never see ads at all.

Who steers the agent?

In tests with Meta's Muse, the agent recommended a robot vacuum purchasable directly within Muse — not the top model in the review it cited. Mark Zuckerberg told Meta's October earnings call the company will "build commerce tools so we get paid a fee based on the transactions that we help facilitate."

Consumer Reports has proposed engineering "loyalty by design" into agents, requiring them to put user needs first and disclose conflicts. Sen. Mark Warner has floated draft legislation treating certain AI agents as fiduciaries under law.

The market may correct bad agents — but switching is costly. An agent that knows your email, calendar and purchase history is far harder to fire than a search engine. Labels protect readers from sketchy sources. Nothing yet invented protects them from disloyal agents.