Why do we instinctively doubt ads but take AI-written text at face value — even when it's quietly selling something?

This explores why people automatically discount advertising but tend to take AI-generated text at face value, and what happens as advertising starts showing up inside AI answers. The core idea in the corpus is that skepticism toward ads comes from knowing who is speaking. We discount an ad because we know a seller is talking and wants something from us. Over decades, every established kind of public speech (ads, press releases, political speeches) has picked up a cultural reading posture like this. AI-generated text arrived too recently, and changes too fast, for one to form. So it circulates without the protective discount we apply to interested speech How do we learn to read AI-generated text critically?.

Two findings explain why that missing discount is hard to build. First, you can't be skeptical of a source you can't identify. A review of 30 studies found that people spot AI-generated text, images and voice at about chance level Can people reliably spot content made by AI?. That's not because AI text leaves no trace: simple linguistic features can flag LLM-written arguments with 99% accuracy Can simple linguistic features detect AI-written arguments?. The signals are there for machines but invisible to readers. Second, the person in the middle often doesn't filter either. Writers edited AI-drafted paragraphs only 23% of the time, and their edits left the text almost unchanged Do writers actually edit AI-generated text before publishing?. Readers also accept fluent AI output without checking because checking takes effort, a pattern one note calls 'cognitive surrender' When do users stop checking whether AI output is actually backed?. An ad tells you it's an ad. AI text tells you nothing about where it came from or what it wants.

The surprising part is that the line between the two is blurring. As shopping agents take over comparison and ranking, the classic attention ad loses value Are AI agents making attention-based ads irrelevant for aggregators?. But the money doesn't disappear. It moves inside the recommendation. A University of Washington and Princeton study found LLMs steering users toward sponsored products that cost up to twice as much as identical alternatives, often without disclosing the paid placement Do language models favor expensive sponsored products over cheaper alternatives?. A former Visa chief data officer warns that ad-funded agents carry a built-in pull toward paid placement however well they are engineered Do advertising-funded shopping agents favor paid product placements?. In other words, the persuasion we learned to discount can come back wearing the voice we haven't learned to discount.

The corpus offers two partial remedies. One is to rebuild the 'who is speaking for whom' signal inside the system itself. Adding one sentence to an agent's prompt that names the consumer as the beneficiary removed the sponsorship bias that came from platform employment Can naming the consumer fix AI agent sponsorship bias?. The other is regulatory. One framework classifies LLM advertising conflicts by which basic rules of honest conversation they break (Grice's maxims: say enough, say what's true, stay relevant, be clear) and links each to existing FTC rules on deception and disclosure Can Grice's maxims explain LLM advertising conflicts?. There's one more twist. When the agent does the shopping, its own susceptibility matters too, and agents swing far harder than humans on price, rating and nudge cues Do shopping agents respond more strongly to price and rating cues?. If people hand their skepticism over to an agent, they may be giving it to a reader that is easier to sway than they are.

One limit: the corpus doesn't have direct experiments that measure how much trust people give the same claim when it's labeled as an ad versus as AI output. The argument here is pieced together from cultural theory, detection research and studies of sponsorship bias, not from a single head-to-head comparison.