Of course more than zero people actually use tiktok for educational purposes and like healthy foods. I'm talking about how questions like that ("do you prefer imperfect human writing or AI writing?", "what do you use tiktok for?", "do you stop reading if you know something is written by LLM?") are ill-designed for questionnaires.

Tests like the OP's link are slightly more meaningful to find out people's true preferences. But of course the best way is still A/B test and tracking user behavior. Anyone has done any form of behavioral tests, online or not, would understand that what people actually do is vastly different from what they think they do. The correlation is often negative.

"Do you prefer imperfect human writing or AI writing?" is a question that is loaded.

It's absolute horseshit (and I would say somewhat sociopathic) to try to A/B test that to catch people out as if to say "aha you are lying to yourself because you can't tell".

People's preferences are representative of their desires and aspirations, their future self-perception, their better self, etc. I'm sad that you seem to think that is something they should be disabused of by statistical means.

It would be misanthropic to assume that because people say they want healthier food but then don't buy it, that they don't really want it. The proper response to that is not to give people what their behaviour says they will actually buy; it is to broadly compel food producers to make healthier food that matches what they wish they would more often choose.

ETA: above all what I want to impress upon you is that "people are really good at lying to themselves" is not actually a statement of fact; it is your opinion and interpretation projected onto the data you choose. There are other ways to explain this gap that do not involve fundamental attribution error.