Hand a draft over to AI and start polishing the sentences, and at some point it gets hard to tell where "the part I actually wrote" ends in the finished piece. A paragraph that was clearly AI-generated at first can start to feel, after a few rounds of editing, as if you wrote it from scratch. Flip it around, and a piece that began as your own idea but was handed to AI purely for wording can end up feeling strangely unfamiliar.

For anyone creating content solo, this mismatch in perception isn't trivial. Whether it's a newsletter, a proposal, or a blog post, deciding "this is my work" is the starting point for figuring out how to write the next piece and what you still need to learn. So what happens when that very judgment becomes shaky? A recent study dug empirically into exactly this question: how accurately people perceive their own contribution to AI-assisted writing.

The question behind the study

Researchers Célina Treuillier and Denis Lalanne argue that as generative AI has spread, it has become urgent to ask exactly how people collaborate when they create new content by interacting with AI. To get at this, they propose a new concept: "authorship calibration."

Put simply, it describes how accurately a person gauges their own real contribution when writing alongside AI. When your actual contribution lines up with your perceived contribution, your authorship perception is well-calibrated; when the two diverge sharply, it's miscalibrated. To measure this, the researchers used the CoAuthor dataset to analyze how people's sense of authorship shifts during AI-assisted writing.

What the study found

The first thing the paper establishes is that authorship perception varies a great deal from person to person. Rather than everyone being similarly mistaken or similarly accurate, the study found pronounced individual differences.

The more striking finding is that this variation tracks how often someone uses AI. People who leaned on AI more heavily tended to misjudge their own contribution, while those who used it less accurately perceived their own share of the work. The researchers interpret this as evidence that AI itself can blur a person's sense of authorship.

The paper also flags implications for learning contexts. Misjudging your own contribution throws off metacognitive monitoring — your ability to track your own thought process — which in turn skews learning strategies and can ultimately affect learning outcomes. The researchers conclude that cultivating accurate authorship perception is central to using AI responsibly and in ways that genuinely support learning.

Applying this to real work

For solo creators, this finding points to one useful habit: before you ship a piece, pause and ask which part of it you actually made.

Turning the study's finding around, if heavier AI use tends to distort authorship perception, then the more you rely on AI, the more this check-in matters. If you handed a draft to AI, note whether the skeleton and the argument were yours, or whether you only refined the wording versus following the AI's structure wholesale. Keeping that distinction clear keeps your sense of contribution to AI-assisted writing closer to reality.

This habit also ties into skill-building. Knowing precisely where you consistently lean on AI lets you choose, deliberately, whether to fill that gap yourself or hand it to the tool on purpose. Someone who honestly tracks their own contribution also learns exactly which muscles they still need to build.

A caveat

That said, it's worth being careful not to over-extend this study. The analysis is based on a specific dataset, CoAuthor, and what emerges at the abstract level is a correlation between usage frequency and authorship perception. It's more accurate to read this as a signal that such a tendency was observed, rather than a firm claim that heavy AI use always leads to misperception.

It's also worth noting that the researchers' focus is primarily on learning contexts. Before mapping this directly onto the everyday writing of a solo entrepreneur, it's worth judging for yourself how much your own work actually overlaps with that setting. Even so, the core observation — that your sense of knowing exactly what you made can itself be unreliable — is worth reflecting on for anyone who exchanges drafts with AI on a daily basis.