The Costly Signal
A blogger closes out a technical post with a line insisting he wrote every bit of it himself, no assistant touched the code, "like a good caveman." The obvious reaction: so what. Anyone can type that sentence whether or not it's true. A disclaimer nobody can check is worth exactly nothing.
That reaction is correct about the sentence and wrong about the job the sentence is doing. It reads like a certification, something with a truth value you could in principle check against a registry. There's no registry. What the sentence is actually doing is putting on a display, and a display works by a different mechanism than a certification does.
A peacock's tail doesn't convince a peahen because she can inspect it against some genetic ledger and confirm the claim. There's no ledger to check. The tail convinces her because growing one that size and dragging it through underbrush without becoming easy prey is an expense only a healthy bird can absorb. The signal is honest because faking it is expensive, not because anyone checked the books.
The disclaimer works the same way, when it works at all. The sentence itself costs nothing to type. What costs something is everything the sentence puts at risk once it's attached to your name in public: every stray inconsistency in the code, every phrase in your prose that reads like default model cadence, now counts as evidence against a specific claim you made about yourself, next to work anyone can go check line by line. Writing "I did this by hand" is free. Writing it and having it hold up under someone actually reading your code is not. That's the cost the peacock is paying. The disclaimer's value comes from the scrutiny it invites, not the words in it.
This is where "curated without AI" badges and organic food labels look identical and turn out to be different mechanisms entirely. Organic certification is also unverifiable to a shopper standing in the aisle, but it isn't a pure display, because behind the label there's an inspection apparatus, paperwork, an entity whose job is to occasionally check. Expensive and imperfect, but real. Nothing like that exists behind a blog's hand-written badge, and nothing like it is coming, because verifying that a specific paragraph was human-typed costs about what writing the paragraph costs. There's no cheaper check to build. That's fine. It was never trying to be a certification.
The confusion runs one direction, consistently. Someone reads the disclaimer, correctly notices no evidence is attached, and concludes the whole ritual is empty performance. But a costless performance wouldn't survive being attached to a post from a writer with an archive, read by people who know his sentences. It survives because it's a bet against your own track record every time you place it, and stating a bet out loud only makes sense if losing it would actually cost you something.
Watermarking schemes for AI text failed for close to the opposite reason. Those tried to build a certification, a checkable stamp enforced by whoever has the incentive to cheat, and any scheme like that collapses the moment the method goes public, because the party being checked simply stops cooperating once it knows how the check works. A costly signal doesn't need the signaler's cooperation in that sense. It needs the signaler to have exposed themselves to a cost a liar can't cheaply match, visible whether or not the audience trusts a word of the accompanying text.
So the genre isn't hollow because disclaimers are inherently theater. It's hollow exactly when the cost isn't there, when the sentence gets appended out of habit with nothing behind it to punish a lie. Read correctly, the tell was never the sentence. It's whether the writer has enough of a public, checkable record that getting caught would mean something. A disclaimer on a first post from an anonymous account is a peacock with no tail claiming one. The same sentence attached to a decade of archived work is a different animal, even though the words are identical.