Stanley Druckenmiller wrote a Wall Street Journal op-ed about bond markets. A detection tool said a machine wrote it. He shrugged, agreed, compared it to using a calculator, and said the message was his. Then a thousand people had the same argument with each other. Most of it was about disclosure. Should he have announced it. Should op-eds come with a bibliography naming which model did what. Is it fair that a billionaire gets a shrug and a first-time author gets dragged. Fine questions. Every one of them aimed about a foot to the left of the thing that matters. I write books for other people. My name is on none of them. That deal is a century old and it already answered this, and the answer has nothing to do with who typed it. The silliest part of the whole fight is how much of it is about keystrokes. AI is a tool, and a writer who treats it like one is doing what writers have always done with help. I’d like to see the same thousand people spend that energy asking whether the op-ed’s argument holds up.
Why doesn’t an AI detector score prove an op-ed was AI-written?
The tool caught structure. One of the tells people pointed at was the move where a sentence denies one thing to assert another. It’s not this, it’s that.
I ban that construction. Been on my scrub list for years, long before anybody worried about machines, because it sounds like insight and delivers none. When I edit a manuscript I hunt it and kill it.
So look at what the detector measured. The detector found a pattern machines overuse. Machines overuse it because humans wrote it into everything the machines learned from. A writer with that habit fails the test. A machine told to avoid it passes. Style detection catches style. That’s the whole ceiling. It can’t get anywhere near the only question worth asking. Does the guy know what he’s talking about. I think it’s reckless for an editor or a publisher to act on a detector score. Those tools flag careful writers and wave through anybody who has learned to prompt around them, and a writer who loses a contract over a false positive has no way to argue with a number.
What ghostwriting already settled
Presidents have speechwriters. CEOs have comms teams. Celebrity memoirs come from people whose names appear nowhere in them. Business books, almost all of them, had help.
Nobody is scandalized, and it’s not because we agreed to pretend. The value in those pages was the judgment underneath, and the judgment belongs to the person on the cover.
A client comes to me with forty years of knowing something and no way to get it onto paper.
I ask questions until the real thing falls out, then I write it so a stranger will read sixty thousand words of it. The book is his. I’ve done that 54+ times and never once confused myself for the author.
I’ve done that 54+ times and never once confused myself for the author.Share on X
Druckenmiller ran the same play with a different assistant. Forty years of trading made the judgment. Something else typed it. Whether the judgment is any good is a real question, and you can answer it by reading the thing.
More on how credit works in Do Ghostwriters Get Credit?
What is the main failure in AI-written op-eds?
The true failure is the guy who has no thought and lets the machine hand him one, and I have no sympathy for him.
Different animal entirely. A ghostwritten book has an author who knows things and a writer who gets it out of him. A generated book has nobody home. A prompt went in, an output came out, and the name on the cover learned nothing and risked nothing.
Readers catch it and can’t say why. They finish and feel like they were talking to a wall. They can’t point at a bad sentence, because every sentence is fine. What’s missing is the thing only that person could have said. The call that went sideways. The number he still remembers. The argument he’s had two hundred times and sharpened every time.
That gap has nothing to do with style. No detector will ever find it. It’s the difference between a book about a subject and a book by somebody who lived it.
What does a client with an AI draft need?
A woman came to me a while back with forty-five thousand words she’d bought from an AI service. A novel out of her own life. Right length, clean grammar, delivered as a file.
She hired a ghostwriter anyway. Smart.
The tool can’t give you the material. It has no idea what happened in that room, what scared her, or what her family will say when they read it. What it wrote is a plausible version of a life it never saw, averaged out of every similar story it had ever been fed. It reads fine, and it isn’t hers.
Some of those words will survive. Most of the book has to come out of interviews, because that’s the only place the material lives. The draft gave her a shape. A shape isn’t a book. That’s the whole argument in practice. The machine is useful, and it’s useful at exactly one point: after you have the thing and need help getting it out. Run it before that and you get something book-shaped with nothing inside.
Should writers disclose AI use?
Sometimes. It depends on what the tool did, and I know that’s a duller answer than either side of the fight wants to hear.
Machine fixed your grammar? Nobody announces a copy editor. Machine organized an argument you brought?
That’s what a ghostwriter does, and a hundred years of practice says say nothing. Machine handed you the argument? Now we have a problem, and the problem isn’t that you kept quiet. The problem is that the piece is hollow.
My full structure is in the disclosure piece.
What I won’t sign up for is a sources list on every article naming which tool did which stage. It sounds rigorous. It measures sweat, and nobody was ever buying sweat. Nobody demands a bibliography for a State of the Union, because everybody understands the president owns the position no matter who typed it.
Who gets forgiven, and who does not
One thing in that pile-on deserves to outlive the news cycle. The same admission that gets a shrug for a billionaire can end a first-time author. The writing has nothing to do with it. That’s real and somebody should say it out loud. But a stricter rule enforced evenly doesn’t fix it, because a rule pointed at the wrong target always falls hardest on whoever has the least protection. Better question fixes it. Did the person have the thought. Fair to a billionaire, fair to a debut author, survives every upgrade to the tools, and any reader can settle it by reading. The debut author is also the person most likely to be writing with AI because professional help costs more than they have. I think that’s fine. Somebody without the money for a ghostwriter who uses the tool to get a book out is doing something sensible, and a disclosure regime built with billionaires in mind would punish exactly that person.
What I tell clients
Use whatever you want. I use these tools every day. One rule. Be in the book. If the argument is yours, if the stories happened to you, if a reader closes it knowing something he could only have gotten from you, then it’s your book no matter what typed it. If none of that is true, no disclosure line saves you, and nobody needed a detector to find the problem.
Druckenmiller has forty years of scars and an opinion about the deficit, and whether he’s right is worth an argument. Whether he typed it himself is the least interesting thing in the story. A reader who acts on a bad bond call can lose real money, and that’s where the thousand people in that argument should have aimed their outrage. I think they wasted a week on the wrong fight.
