Voice Is the One Thing AI Can Approximate But Never Actually Copy

If you've read enough of a writer's work, you can recognize them in a sentence. David Foster Wallace's footnotes-within-footnotes spiral. Joan Didion's declarative sentences that land like a door closing. George Orwell's refusal to let a Latinate word stand where a plain English one would do.

That thing — the thing you recognize before you check the byline — is voice. And it's the one thing AI can convincingly approximate in short bursts but can't sustain over length or reproduce with fidelity.

What voice actually is

Voice isn't style. Style is something you can describe: sentence length, word choice, punctuation habits. Voice is what those choices express about the person making them. It's the attitude behind the syntax.

When you read Didion, the short declarative sentences don't just create a certain rhythm. They communicate something about how she sees the world — with a kind of cold clarity, without sentimentality. The style is the signature of the perception.

AI can reproduce the style. It can generate short declarative sentences. What it can't do is have Didion's actual perception, because that comes from having lived a specific life, having noticed specific things, and having developed specific intellectual obsessions over decades.

Why approximation isn't the same thing

If you ask an AI to write "in the style of" someone, you'll get the mannerisms without the substance. It's like someone doing an impression — you can hear the impression without mistaking it for the original.

In shorter texts, this matters less. A one-paragraph summary doesn't need a distinctive voice to be useful. But the longer a piece gets, the more obvious it becomes that the voice isn't coming from anywhere real. There's no there there.

What this means for writers

The writers who are going to be fine are the ones with something to say. Not a broad topic, but a specific angle, shaped by their particular experience, that nobody else would arrive at in quite the same way.

Generic content — "here are five tips for productivity" — was always a commodity. AI has made it cheaper to produce, which means the market for it has collapsed. This is probably fine. It was always the least interesting part of writing.

The work that mattered before AI still matters. It's just more visible now that the filler is so easy to generate. The readers who care about the difference will find the writers who still deliver it.