People have no mental model for it. They know people don't know much, and are used to computers being deterministic. Now (setting aside complete rubes who believe anything) people are used to detecting human-generated BS and spam on the Internet.

- It's nearly always poorly written because dummies also usually can barely write

- It's poorly packaged (some random comments section, not a polished slide deck)

- It's biased in predictable ways (when you do know who wrote it, their self-interested takes are predictable).

Enter "AI" - it writes mechanically well (even though it overuses several tropes, it doesn't make grammatical errors). AI packages its output as exactly what you ask it to (e.g. finished coding projects, outlines, essays, etc.) And not being a person, it's not guaranteed to have a certain bias, but it does echo and validate all your own biases.

Given all this, people think AI sounds exactly like what they assume an oracle would sound like.