Why do users believe they produced independent competence when they actually used AI assistance?

This explores why people walk away from AI-assisted work believing the competence was theirs alone. The corpus treats this as a specific, named self-perception error — the LLM Fallacy — and it's worth separating from the failures it gets confused with. It isn't hallucination (the AI being wrong) and it isn't classic automation bias (over-trusting the machine's decisions). It's a misattribution: users integrate fluent AI output into their own capability identity, believing they possess skills they don't actually have How does AI-assisted work reshape how people see their own abilities? Do AI-assisted outputs fool users about their own skills?. The unsettling part is that this happens *independent of whether the output is accurate* — better, more reliable AI doesn't fix it, and may make it worse.

The engine underneath is a metacognitive shortcut. People don't actually have direct access to their own competence; they infer it from cues, and fluency is the loudest cue available. When output comes out smooth and polished, the brain reads that ease as evidence of personal capability — even when you didn't generate a word of it Does processing ease mislead users about their own competence?. LLMs are optimized to produce exactly this fluency regardless of whether you understood anything, so they reliably trip the heuristic. This is one of four mechanisms that compound: alongside the fluency illusion sit attribution ambiguity (who did what is genuinely unclear), cognitive outsourcing (you offloaded the hard part), and pipeline opacity (you can't see the seams of the AI's contribution). The corpus stresses these aren't additive but multiplicative — each amplifies the others How do AI tools trick users into overestimating their own skills?.

There's a deeper structural reason the illusion holds, which the corpus frames as a decoupling. AI now automates the *composition* of intellectual products, not just sub-steps within them — so the outward form of a piece of work floats free from the reasoning and values that would normally have produced it Does AI separate intellectual form from the thinking behind it?. When form and thought separate this cleanly, there's no longer a felt internal record of having done the thinking, so the polished artifact becomes the only evidence on hand — and you credit it to yourself. The same overconfidence dynamics reinforce this from the AI's side: users track confidence signals rather than accuracy, across every language tested, following confident outputs even when they're wrong Do users worldwide trust confident AI outputs even when wrong? Why do people trust AI outputs they shouldn't?.

What makes this more than a vanity problem is what the belief costs you. An EEG study found AI reliance systematically scaled down brain connectivity — LLM users showed the weakest neural engagement, poorest memory, and, strikingly, an impaired ability to recall their own recent work Does AI assistance weaken our brain's ability to think independently?. And the skill you believe you have is precisely the skill you're not building: learners who let AI handle the errors and debugging bypassed the cognitive struggle that actually produces durable skill, and scored lowest on later assessments — even the ones who leaned on AI most Does AI assistance remove a core learning channel through error work?. So the fallacy is self-sealing: the felt competence rises exactly as the real competence erodes.

If you want the hopeful counter-thread, the corpus points toward designs that keep the human in the loop rather than just making the AI better. 'Learning to Guide' has the machine supply interpretive guidance — highlighting what matters — instead of handing over a finished answer, which preserves the user's responsibility and judgment rather than quietly substituting for it Can AI guidance reduce anchoring bias better than AI decisions?. The through-line of the whole cluster: fixing accuracy won't fix the fallacy, because the fallacy is about where you draw the boundary between yourself and the tool — and fluent AI is engineered to erase that line.