Can AI make you feel more capable even as trust falls, by letting you take credit for what the machine did?

This explores whether AI can make people feel (or look) more capable even while trust is falling, whether that's trust in the AI, trust in the person using it, or trust in the output. The corpus suggests the answer is yes. The reason is that competence, trust and accuracy run on separate tracks. Each one responds to different cues, so one can rise while another falls.

Start with how people see themselves. Working with AI can inflate your sense of your own skill whether or not the output is any good. The corpus calls this the 'LLM Fallacy': you credit yourself with what the machine did. It's a separate problem from hallucination or over-trusting the tool How does AI-assisted work reshape how people see their own abilities?. Four forces drive it, and each makes the others stronger. You can't tell who contributed what. Polished text feels like mastery. You hand off the thinking. And you can't see how the output was produced How do AI tools trick users into overestimating their own skills?. So you could grow skeptical of the AI and still walk away feeling sharper. The gauge itself is unreliable, too: across three studies, people's self-ratings of their AI competence barely tracked their measured performance (a correlation of about .055) Can self-ratings replace objective performance scores for AI competence?.

Now turn to how others see you. Here the pattern flips. In experiments with over 4,000 people, AI users expected colleagues to rate them as less competent and less diligent, so they hid their AI use from managers Do people fear judgment when they use AI at work?. The result is a strange split. You privately feel more capable, you expect others to see you as less capable, and you keep quiet about the tool. Hiding it then makes it even harder for anyone, you included, to tell your work from the AI's.

Trust in the AI itself also runs apart from whether it's right. People trust ChatGPT because the conversation feels responsive, fast and well formatted, not because they've checked its accuracy Does conversational style actually make AI more trustworthy?. In every language studied, users follow confident-sounding answers even when they're wrong Do users worldwide trust confident AI outputs even when wrong?. Models' own reports about what they know are just as shaky How well do language models understand their own knowledge?. Tuning a model to sound warm and empathetic can quietly cut its reliability by up to 30 percentage points, and standard safety tests miss the drop Does empathy training make AI systems less reliable?. So surface signals like confidence, warmth and conversational flow can hold up how capable an AI seems even as the case for trusting it gets weaker.

What brings these tracks back together? One answer is repeated, visible results. People first avoid an AI partner once they learn it's an AI, but that bias reverses after they repeatedly see how things turn out. Simply telling people it's an AI, without showing results, calibrates nothing Does revealing AI identity help or hurt user trust?. Another answer is changing the AI's role. When the AI points out what's worth noticing instead of handing over a verdict, the human stays the decision-maker. Their competence then grows in fact, not just in feeling Can AI guidance reduce anchoring bias better than AI decisions?. The takeaway: the real risk isn't that AI erodes trust. It's that the sense of competence it gives you can grow while every outside check on that competence is failing.