Signing your name to a piece of writing doesn't prove you thought it through, and AI makes that gap easier to hide.

This explores why putting your name on a piece of writing ("I wrote this and I stand behind it") can come apart from how much you actually thought it through, with AI assistance as the sharpest case. The corpus has no note on authorship as a concept. It does cover both sides of the gap: what readers use to judge who wrote something, and what only real cognitive work leaves behind.

On the social side, authorship is read off cheap signals. In a study of 2,939 writers and 11,091 readers, AI assistance shifted how the writer came across on all 29 measured dimensions, making them seem more confident, higher quality and more privileged Does AI writing assistance change how readers perceive the writer?. The picture of who is behind the text was changed by the tool, not by the writer's thinking. Readers also rarely check. Users trust answers with more citations almost as much when the citations are irrelevant as when they're relevant Do users trust citations more when there are simply more of them?. Presuppositions slip new claims past scrutiny by presenting them as settled background Why are presuppositions more persuasive than direct assertions?. What readers already believe predicts persuasion better than what the text says Does what readers believe matter more than what debaters say?. When audiences run on shortcuts, a claim of authorship can stand with nothing behind it and nobody notices.

Reputation is what normally keeps such a claim honest. An expert argument carries force because of the thinker's track record and standing, and that social context is exactly what a text-only model can't see Can language models distinguish expert arguments from common assumptions?. AI-generated posts collect social proof without building any speaker's sustained reputation ai-displaces-influencer-content-threatens-social-medias-social-proof-function. The credit accrues, but the accountability ledger doesn't. Without reputation at stake, a byline is just a label.

The gap still leaks, because real thinking leaves traces that polish can't add. A classifier separated AI from human fiction 93.2% of the time using only story-level choices like character agency and chronological structure. It kept 97% of its accuracy with style cues removed, and those choices resist disguise because they take rewrites, not surface edits Can AI stories be detected without analyzing writing style?. Readers describe AI posts as aloof, and one explanation is that human writing makes an internal appeal to the reader's attention that AI text doesn't perform Does AI writing lack the internal appeal to attention that humans use?. Engagement shows up in structure and in address to a reader, which is why editing the surface can't fake it.

Machines show the same split. Deep research agents invent examples, products and evidence to look scholarly when depth is demanded, and this accounts for 39% of the failures analyzed Why do deep research agents fabricate scholarly content?. They perform the signals of rigor without doing the work. Research writing also looks like iterative draft-and-revise, with a persistent draft repeatedly cleaned up through targeted retrieval Can iterative revision cycles match how humans actually write?. If so, the thinking lives in the revision loop, and putting your name on a draft you only accepted claims a process you never ran. That last step is my inference, not a finding from the note. The divergence exists because the claim is judged by cheap signals nobody audits, while engagement leaves costly traces that almost nobody checks for.