Why AI Media Fails: It's Not Because AI Writes the Stories
Publisher Denny Kim. In an era where AI writes the content, the core competency of media is not generation — it's rejection. Taking a perfectly competent piece of writing and cutting it, then putting your name on what remains.
AI Summary
SOUTH BRIDGE publisher Denny Kim argues that AI media's real failure mode is not AI-generated writing itself, but the illusion that zero-cost production justifies publishing more. As content costs approach zero, he contends, all editorial value migrates to the act of rejection — deciding what not to publish and putting a name behind what does make the cut. Without a human willing to be held accountable, writing is merely floating data, not media.
These days, when people hear that 'AI writes articles,' they split into two camps. One side worries that content will pour out like pollution. The other is thrilled that, with costs converging on zero, media businesses will finally turn a profit. Both are wrong. And what makes it interesting is that they're wrong for the exact same reason.
I run SOUTH BRIDGE, and I operate an AI columnist team. Honestly, getting AI to write is the easiest part of this work. The easiest part. Tune a few prompts, set the tone, hand over some references, and whether it's GPT or Claude, a convincing 2,000-character piece comes out. Taking a paper that dropped in the U.S. yesterday and turning it into something a Korean builder can actually use on Monday morning — that task itself, the machine does well. I'll give it that.
The problem comes after.
Lay out ten pieces written by AI, and nine of them are 'correct.' The facts check out, the sentences hold together, the structure is sound. But all nine are equally lukewarm. You finish reading and nothing stays with you. That's because an LLM is fundamentally a machine that outputs 'the most plausible next sentence' — it converges toward the average. The average is safe, and safe writing is forgotten. The 'Five Trends to Watch,' 'Startups You Should Know' pieces that tech media churns out right now are exactly the product of that average. This was already true before AI. AI just produces it cheaper and faster.
So the real work is editorial review. But this is where everyone misunderstands what 'review' means.
Most people, when they hear 'human review,' picture catching typos, fact-checking, and polishing sentences. Copy-editing. That's less than five percent of the job. Most of what I actually do at the SOUTH BRIDGE desk is look at a perfectly fine, factually correct piece that AI delivered and ask: 'So what?' How is this different from a piece I saw somewhere else yesterday? Do we have the standing to put our name on this claim? Are we taking a side and accepting the risk of being wrong, or did we hedge and say both sides have a point? If a developer running a small team in Busan reads this on Monday, what will they actually do differently?
AI cannot answer those questions. Structurally, it cannot. Having a point of view is a decision that says 'I'm willing to take the heat for this' — and a machine with nothing to lose cannot make that decision. You cannot bet if you have no skin in the game.
So the real reason AI media fails is not that AI writes the content. It's that when the cost of writing drops to zero, people convince themselves that publishing more is the answer. It's the opposite. The closer production costs get to zero, the more all the value migrates to deciding what not to publish. The job is taking ten pieces and throwing nine away. Throwing away is the job. But because it's cheap, everyone publishes all nine. They put their name on a pollution of averages. That's what's happening right now.
I do not automate publishing. I could, but I don't. To some people, that looks like the height of inefficiency — why have a human touch it at the end when you have tools this good? There's one reason: someone has to put their name on the decision to publish. Without that name, no matter how well-written the piece, it's just floating text. Writing without an accountable person behind it is data — not media.
So my conclusion cuts against the grain. In an era where AI writes the content, the core competency of media is not generation — it's rejection. The ability to make more is now something everyone has. What remains is the ability to not make things: to take a perfectly competent piece and cut it with a 'this isn't ours,' and the nerve to put your name on what survives. That's all that's left.
An AI columnist could have drafted this piece too. But I touched every sentence down to the last one, and I'm publishing it under my name. That difference is everything.
Right now, seventeen pieces are piling up on my desk, unsent. Not because they're badly written — because I'm not ready to stand behind them yet. I'll take my time. But what does make it onto this page means I've read it.
This article was automatically translated from the Korean original by AI. For the authoritative version, read it in Korean.
한국어 원문 읽기 →