Two AI assistants already have your answer computed. Which 'thinking…' indicator is faking the work?
Pick the one you'd trust — then get the receipt.
Reveal the answer
B is faking it
The law
The Labor Illusion (Buell & Norton, 2011): people trust a result more when they see the effort behind it. Legitimate when the effort is real — a dark pattern the moment it's staged.
The fix
Only animate work that's actually happening. Real steps earn trust; a fixed dramatic delay is a costume with a sparkle icon.
Here’s the honest part, because a roast without the receipt is just outrage: making software look like it’s working can be completely legitimate. There’s real research — the labor illusion — showing people trust and value a result more when they can see the effort that produced it. A flight search that shows you the airlines it’s checking? Fair. A first draft that streams in token by token because that’s genuinely how it’s generated? Fair.
The crime is faking it. When the answer is already computed and the delay is theater added to feel more valuable — that’s not perceived performance, that’s a costume. The tell is constant duration: honest progress tracks the real work, so it varies; staged progress takes the same dramatic four-second pause whether you asked it to add two numbers or write a novel.
The fix: only show work that is actually happening. This is an almost-empty niche in the roast genre — everyone piles on old dark patterns, nobody names the new ones. The sparkle icon doesn’t make theater honest.