Orlean on her first-book-deal failure: "There's a stench of inauthenticity that is very easy to smell on people."1 Eleven publishers passed on her triathlon proposal. Their rejection letters praised the writer ("She seems terrific") and declined the project. The proposal was competent. The writer was capable. What was missing was the I'm dying to do this — and somehow eleven different publishers detected the absence.
The proposal was the same paragraphs of prose either way. What was the stench operating on? Word choice density? Time spent on different sections? Energy distribution? Something below conscious analysis that professional readers had trained themselves to detect over thousands of proposals.
The wider claim: markets price for authenticity-signal everywhere they can detect it, and the detection works through proxies the seller cannot performatively construct.
A few examples:
The cross-cutting question: what are the universal proxies for authenticity-signal? My current guess: (1) specificity-density (authentic creators reach for specific examples; inauthentic ones default to general claims); (2) friction-pattern (authentic creators dwell on the difficult parts; inauthentic ones brush past them); (3) attention-distribution (authentic creators give more attention to the part that surprised them; inauthentic ones distribute attention by surface logic).
If these proxies are real, they suggest a writing practice: protect the friction, dwell on the surprise, refuse to flatten the specific. Anything that performs authenticity by faking smoothness reads as inauthentic. Anything that retains the actual messy texture of the writer's encounter with the material reads as authentic.
The deeper question: AI text generation has the option to mimic the texture — to fake friction, fake specificity, fake dwell. As it gets better at this, the proxies stop discriminating. What's left? The candidate proxy I haven't been able to falsify: what the writer would have written if no audience were watching. The audience-blind text has different attention-distribution than the audience-aware text. AI does not yet generate audience-blind text reliably. This may be the deepest authenticity-proxy and the last one to fall.