Tell people an AI wrote the message, and they scrutinize it harder, yet many still end up convinced.

This explores whether telling people that AI wrote a message protects them, or whether they get more skeptical and still end up convinced. The short answer from the corpus is that the second pattern is real. In one study, people who knew AI was involved scrutinized the message more closely, but between 34% and 62% of them were still persuaded, depending on the group Does telling people an AI wrote something actually stop them from believing it?. Disclosure switches on critical thinking without switching off the message's pull. It helps, but it isn't enough to work as a safety mechanism on its own.

The surprising part is that the *kind* of warning seems to matter more than the fact of disclosure. A different line of experiments, with 3,208 Americans, didn't just say "this is AI." It warned people that LLMs *can be prompted to persuade*, and that cut belief change roughly in half without making people distrust AI in general Can a simple warning reduce how much LLMs persuade people?. One way to read the gap: "an AI wrote this" tells you about the source, while "this could be built to change your mind" tells you about the motive. One essay in the collection argues we already discount advertising automatically because we know it's interested speech, but AI-generated text arrived too fast for us to build a similar reflex How do we learn to read AI-generated text critically?. A persuasion warning may work because it lends people that missing frame for a moment.

It also helps to separate *trust* from *persuasion*, because disclosure moves them differently. Revealing AI authorship reliably lowers how trustworthy, caring and likable readers find the writer, and the drop is steepest in personal, interpersonal writing How does revealing AI authorship change reader trust?. Readers also want disclosure more than writers think is necessary Do readers and writers differ on AI disclosure necessity?. But liking the messenger less is not the same as rejecting the message. AI assistance also quietly changes how a writer comes across, making them seem more confident, more extreme and more polished on all 29 traits one study measured Does AI writing assistance change how readers perceive the writer?. Those are the very qualities that make an argument land, whether or not the reader is suspicious of the source.

There are also reasons to doubt that the scrutiny lasts. When people repeatedly see an AI partner deliver good outcomes, their initial bias against it reverses Does revealing AI identity help or hurt user trust?. A disclosure label may lose its effect with familiarity. Separately, people still react to an AI as a social presence when it has even one strong cue, like a voice or a face, which suggests that knowing it's a machine doesn't stop us from responding to it as someone Do more social cues always make AI feel more present?. At scale, careful checking may simply fail to keep up with the volume of AI output Can AI generate knowledge faster than humans can evaluate it?.

The corpus doesn't directly test whether the 'more critical but still persuaded' state persists over weeks, or whether combining disclosure with a warning about motive adds up to more than either alone. Those look like the open questions. The most useful lesson for now: labeling a message as AI-made is a weaker defense than telling people why it might be trying to change their minds.