A headline in a tech newsletter caught my eye recently. It was about Amazon’s new AI-based recruiting system and a design philosophy called “humorphism” — the idea of adapting technology to human workflows rather than forcing humans to adapt to technology. I like the focus on human-first in an AI world, maybe that’s deserving of its own post…
But first I wanted to find out more about this tool and Amazon’s approach.
What I found when I went looking
I searched for other coverage of the topic. The first result I found was a clearly AI-generated summary that added nothing. It was a rehash of the same thin set of information: nothing new, no real analysis. I moved on.
Another article I found had something I hadn’t seen before in place of a byline: “Reviewed by [name].”
My first reaction was positive. I thought “at least they’re not pretending a human wrote it.” Obfuscating purely AI-generated content by attaching a human name as author, or just leaving off a byline in the hope people won’t notice, isn’t working and it seems the industry knows it. This at least looked like a more honest approach.
Then I read the article.
(I’m not linking to any of these – they’re not the point. Examples of this pattern are all over the internet, and my intent isn’t to call out specific publishers but to talk about the pattern.)
The catch-22 they created for themselves
The piece wasn’t good. Despite the reported human review it smelled of AI slop. Whatever review happened, it didn’t seem to involve much human effort or adjustment to what the AI generated. It read like every other AI-generated article I see that seems purpose-built to recycle existing information as a mechanism for generating ad impressions.
That’s the problem with the label. By calling the article “reviewed by a human,” the publisher raised the bar for what readers should expect. It told me a human looked at this and decided it was worth publishing. That implies a standard of quality that the article didn’t meet. Now if (when) the article still reads like near-slop, the human reviewer is on the hook for that. The label makes the AI-generated-ness of the content more obvious and bothersome, and makes the human reviewer look bad for approving it.
By publishing primarily AI-generated content they’ve put themselves in a catch-22. They either acknowledge the AI and raise expectations, or don’t acknowledge it and readers notice anyway. There’s no clean way out once you’ve made the choice to deliver poor quality AI-generated content.
This is the world we were warned of, and it’s not going well
A year or two ago, experts began predicting that AI-generated content would flood the internet, get used to train future AI models, and create a feedback loop of degrading quality. That prediction has sadly arrived.
What’s made it worse is lazy human oversight. The “reviewed by” label is a symptom of that. It suggests a human in the loop, but in practice it’s a human who isn’t really engaged. Access to AI is making it easy to publish more with less effort, and the temptation to let the AI do the work and just rubber-stamp it is hard to resist. As that becomes more and more the norm, the loop tightens: more AI content, trained on more AI content, reviewed with less and less genuine human judgment.
The snake is eating its tail.
What “good” actually looks like
There are two honest paths forward for publishers using AI.
The first is to own it: be transparent that content is AI-generated, accept that some readers will discount it, and live with that trade-off. At least it’s honest.
The second is harder but better: partner with AI well enough that the human voice, judgment, and expertise are unmistakably primary. The result is a human-created product that used AI as a tool. In the same way that journalists don’t disclose brainstorm partners, reviewers, etc, who helped them get to their final published article, a piece written by a human that used AI to help them refine their thoughts doesn’t need to be disclosed or excused.
The “reviewed by” convention tries to thread a needle between those two paths and ends up in the worst of both worlds: acknowledged as AI, and not good enough to justify the human stamp of approval.
As I wrote in a previous post “The real story of vibe coding isn’t being told,” AI doesn’t replace human judgment. It just makes it more obvious when human judgment is missing.