The Economist published a piece this week with a thesis that should make AI doomers and AI cheerleaders equally uncomfortable: bots are often bad writers, but so are most humans.
The argument, tied to a recent controversy over AI involvement in a novel, cuts through the usual binary debate. One camp insists AI will never write anything worth reading. The other insists it already writes better than people. The Economist's position is more honest and more awkward: the floor for AI writing and the floor for human writing are closer together than either side wants to admit.
This tracks with what anyone who has used ChatGPT or Claude for long-form writing already knows. The output is competent, structured, and relentlessly mid. It reads like a solid B-minus college essay or a corporate blog post written by someone who read the brief but has no opinion about it. The problem isn't that AI writes poorly in a technical sense. Grammar is fine. Transitions work. The problem is that it writes without stakes, without taste, without the willingness to be wrong that makes good prose worth reading.
But here's the part that stings: pick up most self-published books, most marketing copy, most business writing. It reads the same way. Competent. Lifeless. Correct enough to pass, not good enough to remember. The vast majority of professional writing was already operating at the level AI now automates.
What this means practically is that AI writing tools will keep eating the bottom of the market. Reports nobody wanted to write. Summaries nobody wanted to read. Product descriptions that exist to fill a CMS field. That work is going away, and honestly, it was never the part of writing that mattered.
The part that matters, the part where a human voice says something only that particular human could say, remains stubbornly hard to automate. Not because AI lacks capability, but because it lacks a reason to care. That gap might narrow. It hasn't yet.