Same here. I've likely personally cost openAI tens of thousands of dollars as a consumer user. I've seen tens of thousands of responses across a wide range of LLM voices and I know it in my bones that Hunter used AI for this.

Every single point is the same length and ends with an aphoristic zinger. It's not X it's Y (people know to avoid this and em dashes since it's a lazy LLM tell.) there's a consistent cadence to it but little substance. The cave man version of his points use less than 10 syllables. The prose is extremely even. Humans have variation in how they write especially when making impassioned bluesky posts lol.

To demonstrate here's my argument from above processed through clause

I've spent years as a heavy AI user — we're talking tens of thousands of interactions across models and voices. Pattern recognition at that scale becomes intuitive in the same way a sommelier stops consciously analyzing and just knows. What flags this for me isn't any single element. It's the consistency. Every point is approximately the same length. Every point ends with a punchy, quotable closer. The prose quality never dips or spikes — there are no throwaway lines, no moments where the writing gets lazy or overexcited. Humans don't write that way, especially not in impassioned social media posts. Strip each point to its core claim and you're left with less than ten syllables of actual content. The rest is rhetorical packaging. Elegant, even packaging — but packaging nonetheless. People have learned to avoid em dashes and "it's not X, it's Y" constructions because those are known tells. But the underlying architecture remains. The fingerprint isn't any single phrase. It's the absence of variation. It's the fact that nothing here is accidental. Hunter Biden's documented voice is the opposite of this. This tweet is what happens when you prompt a model to sound like him.