You Used AI to Write It. Now You Can't Tell If the Thinking Was Yours.

I finished a draft last month, read it back, and had a genuinely strange thought: did I write this, or did I approve it? Nothing in the sentences was wrong. The argument was mine, the structure was mine, and I could reconstruct the reasoning behind every claim if you asked me to. But I'd used an AI assistant to get from outline to prose, and somewhere in that process the feeling of ownership had gone thin. Not absent — thin. Present enough that I couldn't call the piece someone else's, missing enough that I couldn't quite call it fully mine either.

That specific discomfort has a name now. A 2026 paper in PLOS Mental Health by researcher A.D. Brown, "The Age of Authenticity Anxiety: Artificial Intelligence and Emerging Questions for Mental Health," identifies a distinct psychological phenomenon: chronic uncertainty about whether one's own AI-assisted output remains authentically one's own, paired with a broader difficulty trusting whether the information a person receives is human-generated or algorithmic. This is not the same anxiety that's dominated the discourse since 2023. It isn't "will this technology replace me." It's quieter and stranger — a doubt about authorship that persists even when the job is secure and the work is good.

This Isn't Imposter Syndrome, and Conflating Them Loses the Real Mechanism

The instinct is to file authenticity anxiety under imposter syndrome and move on, because both involve doubting whether you deserve credit for your own output. But the mechanisms are different enough that treating them the same actively gets in the way of addressing either one. Imposter syndrome is a miscalibration between actual competence and self-assessed competence — you did the work, the work is good, and your internal model of your own ability hasn't caught up to the evidence. The fix, insofar as there is one, involves recalibrating that internal model against reality.

Authenticity anxiety doesn't have that shape, because the question it raises isn't "am I good enough to have done this." It's "did I do this at all, in the sense that matters." You're not underestimating your competence. You're genuinely unsure where the boundary sits between your thinking and the tool's output, because for the first time in the history of knowledge work, that boundary is porous by design. A calculator never raised this question, because nobody confused arithmetic with thought. A spell-checker never raised it, because correction isn't generation. A large language model completing your sentence, restructuring your argument, or drafting the paragraph you were about to write raises it constantly, because the thing it's doing looks enough like thinking that the line between "my idea, machine-assisted" and "machine's idea, me-approved" stops being visible from the inside.

The Doubt Doesn't Scale With How Much You Used the Tool

Here's the part that surprised me once I started paying attention to it in my own work: the anxiety doesn't track cleanly with how much AI assistance was actually involved. A paragraph I wrote entirely myself, after reading fifteen AI-generated options and rejecting all of them, can trigger more authorship doubt than a paragraph the model drafted and I edited heavily. That's backwards if you think the anxiety is really about the percentage of words that came from where. It makes sense once you realize the anxiety is tracking something else: not word-origin, but whether the decision-making process itself still feels legible to you afterward.

When I write unassisted, I can usually reconstruct why I chose one sentence over another, because I made that choice in real time and the reasoning left a trace I can follow back. When fifteen AI-generated options sit in front of me and I pick the closest one, then edit it into something I'd actually say, the reasoning trace gets harder to reconstruct — not because I didn't reason, but because the reasoning happened in a comparison-and-rejection mode rather than a generation mode, and comparison-and-rejection doesn't leave the same kind of memory. This matches what Brown's framework points toward: the anxiety correlates less with output composition and more with process opacity — how well a person can account, after the fact, for how a piece of work came to exist in its final form.

The Wrong Fix Is Using the Tool Less. The Right Fix Is Tracking a Different Signal.

The obvious response to authenticity anxiety is to use AI assistance less, on the theory that less involvement means less ambiguity about ownership. In practice this doesn't resolve the anxiety — it just relocates it, because the ambiguity was never really about frequency of use. Someone who avoids AI assistance entirely can still stare at their own unassisted paragraph and wonder if it's actually original or just a recombination of everything they've read, which is a version of the same authorship doubt that predates any of this technology and was simply easier to ignore before a tool made the boundary question unavoidable.

What actually helps, based on both the process-opacity framing and my own experience testing it, is tracking a signal other than word-origin: can you reconstruct and defend the reasoning behind the final version, regardless of which parts you typed versus generated versus selected? If you can walk someone through why the argument is structured the way it is, why you rejected the alternatives, and what you'd change if a specific premise turned out to be wrong — that's authorship, in the sense that actually matters for accountability and quality, independent of which keys got pressed. If you can't reconstruct that reasoning, the problem isn't that AI was involved. It's that you approved something without doing the part of the work that makes approval meaningful, and that would have been true with a human ghostwriter fifty years ago too.

So Actually — the Question Was Never "Whose Sentence Is This"

I think the reason authenticity anxiety spreads so easily right now is that it's disguised as a technology question when it's actually an accountability question that AI just made impossible to avoid. "Did I write this" was always a slightly imprecise way of asking "can I stand behind this, explain it, and be responsible for what's wrong with it if something is." Pre-AI, those two questions collapsed into each other often enough that nobody had to separate them. Now they don't collapse automatically anymore, and the discomfort people are feeling is the discomfort of a shortcut disappearing, not evidence that something has actually gone wrong with their capacity to think.

The same collapse-then-separation pattern shows up in how burnout gets misdiagnosed — a familiar-feeling problem turns out, on inspection, to be a different mechanism wearing the old symptom's clothes. Authenticity anxiety isn't a sign that AI has made original thought obsolete. It's a sign that "original thought" was always doing more conceptual work than one word could carry, and the tool just forced the split into view. The question worth sitting with isn't whether you wrote it. It's whether you could still explain it, out loud, to someone who's allowed to ask you why.