If you write for a living, or even seriously for an audience, the next eighteen months will probably tell you whether what you do is something a machine can do for free. That’s the stake. Not whether AI can write — it can — but whether the specific thing you’ve been doing all these years was the kind of writing that needed you, specifically, to do it.
The blogs that survived the AI shift had something in common, and it wasn’t what most people assumed it would be. It wasn’t technical SEO. It wasn’t a clever pivot to video. It wasn’t even community size. The ones still standing — actually read, actually shared, actually trusted — belong to writers who had a particular and slightly weird relationship with their subject. Something specific enough that a language model, drawing on the average of everything ever written, couldn’t have arrived at it by chance.
Most of the conversation about AI and writing has been about defense. How to prove you’re human. How to add the personal touch. How to write in a way that algorithms can’t easily replicate. That framing misses what’s actually happening. The question isn’t whether you can sound human. Models can sound human. The question is whether what you have to say required you, specifically, to have lived through something to say it.
And most of us, if we’re honest, are still figuring out whether the answer is yes.
The conventional wisdom got it backwards
The popular advice during the first wave of AI panic in 2023 and 2024 ran something like this: lean harder into expertise, niche down, build authority, optimize for E-E-A-T. The assumption was that AI would commoditize generalist content and that survival meant becoming more specialized, more credentialed, more verifiably expert.
What actually happened was stranger. A lot of highly credentialed, well-optimized blogs got flattened anyway, because their expertise was the kind a model could approximate within a year. Meanwhile, smaller blogs written by people with no particular authority — except the authority of having thought about something for a very long time in their own peculiar way — kept their readers. A recent philosophical study on generative AI argues that the technology functions less as a threat to creativity and more as a mirror reflecting how much of our existing creative output was already imitative, already averaged, already the kind of thing a machine could synthesize from patterns. The blogs that died were the ones the mirror exposed. The blogs that survived were the ones the mirror couldn’t quite reproduce.
This is harder than it sounds. Most writers, including writers who think of themselves as original, are working from a stack of influences that any reasonably trained model has also read. If your blog is essentially a remix of ten popular books in your field, plus your particular sequence of arrangement, a model can do that. The question is whether there’s something underneath the remix.
What can’t be generated by accident
There’s a useful test. If you handed a transcript of your blog to someone who knew you well — not a fan, but someone who has watched you live — would they recognize you in it? Not just your sentence rhythms. Your obsessions. The specific things you keep circling back to that don’t quite make sense to anyone but you. The grudges you can’t let go of. The hill you’ll die on that nobody asked you to die on.
That’s the thing a model can’t generate by accident. Not because models can’t write idiosyncratically — they can, when prompted — but because they have no skin in the game. They don’t have a father who turns off the heater when he leaves the room and calculates the price of his coffee like he’s still earning warehouse wages. They don’t have a specific memory of being twenty-four and unable to sleep because the loading dock job felt like a small death every morning. They don’t have the particular shape of grief that comes from realizing the version of yourself everyone liked was the version that asked for nothing.
I’ve written before about the loneliness of being known only as the agreeable version of yourself, and the writing equivalent of that pattern is everywhere right now. Bloggers who built audiences by being broadly likeable, broadly useful, broadly competent — the writing version of warm and undemanding — are discovering that broad likeability is exactly what models do best. The version of you that asks nothing of the reader is the version a model can replicate without effort. Closeness, in writing as in friendship, gets built in the rooms where someone is needed for something specific.
The writers who held their readership through this shift tend to share a few dispositional habits worth naming directly. They write about a narrow set of preoccupations from many angles rather than a wide set of topics from one angle. They have a relationship with their subject that predates their blog and would persist if the blog disappeared. They’re willing to be wrong in public — not contrarian for sport, but actually mistaken, actually working something out, actually changing their minds in front of you. Models can simulate uncertainty but can’t actually be uncertain. There’s no stake. A writer who admits, halfway through a piece, that they’re not sure what they think — and then keeps writing toward something — is doing something that requires a self to be at risk. And they have, almost without exception, a tolerance for not being understood right away. The internet trained writers to optimize for instant comprehension, the dopamine-friendly hook, the clear takeaway. Models are now better at instant comprehension than most humans. What’s left for human writers is the kind of writing that asks the reader to slow down, that doesn’t deliver its full meaning on first read, that rewards a second pass.
The imposter problem is real, and it’s not what you think
Most writers I know are quietly working through a version of impostor syndrome that didn’t exist three years ago. Forbes has called this the new AI-driven impostor syndrome — the suspicion that what you do for a living could be done, perhaps already is being done, by a machine that doesn’t need lunch breaks. Psychology Today defines impostor syndrome as the persistent inability to attribute one’s achievements to personal merit, with success being credited instead to luck or external factors. The AI version adds a twist: now there’s a plausible external factor that wasn’t there before. Maybe your readers came because of the algorithm. Maybe your last viral piece was just well-timed. Maybe everything you do could be approximated by a sufficiently large model.
The honest answer, for most of us, is that some of it could be. The competent middle of our work — the explainer paragraphs, the listicle scaffolding, the SEO-friendly intros — yes, a model could have written that. The question worth sitting with is whether there’s something underneath the competent middle. Whether you have a specific obsession or a specific way of seeing that makes the competent middle just the delivery mechanism for something that’s actually yours.
Some writers do. Many don’t. And the ones who don’t aren’t bad writers — they’re writers whose careers were built on a particular phase of the internet that rewarded volume, optimization, and broad applicability. Those things aren’t disappearing, but they’re no longer enough.
The history we keep forgetting
None of this is the first time the blogosphere has gone through a culling event. Writers on this site have covered the rise and fall of MSN Spaces, the strange collapse of Yahoo 360, and the slow disintegration of the early blog networks that once felt permanent. Each shift produced the same pattern: writers who had built their identities around the platform got swept under, and writers who had built their identities around their actual work moved on and kept writing somewhere else.
The AI shift is bigger than any of those, but the underlying mechanism is the same. The platform changes. The infrastructure changes. The economics change. What persists is whether the writer had something that wasn’t dependent on the infrastructure. The bloggers who survived the death of Technorati didn’t survive because they were on Technorati. They survived because Technorati was incidental to what they were actually doing.
The honest question
Here’s where it gets uncomfortable. Most of us, when we ask whether we have something to say that an AI couldn’t generate by accident, want the answer to be yes. We want to believe our voice is distinctive, our perspective is irreducible, our particular combination of experience and observation produces something nobody else could produce.
Sometimes that’s true. Often it’s a little less true than we think. The writing I’m proudest of is the writing where I had no choice but to write it that way — where the subject was tangled up in something specific in my own life, and the only way to get it down was to follow the tangle. The writing I’m least proud of, looking back, is the writing where I was performing a competent version of someone in my niche. A model could have done it. In some sense, a model already had — I was just averaging the same sources, in a slightly different order, with my name on top.
The shift from one kind of writing to the other isn’t a productivity hack or a positioning exercise. Research on intrinsic motivation suggests that the work people do because the activity itself matters to them tends to be both more sustainable and more distinctive than work done for external reward. The writers who are surviving this transition mostly weren’t writing for the algorithm to begin with. They were writing because something in them needed to be worked out on the page, and the audience came as a side effect. The writers who were writing primarily for the audience are now competing with a system that can produce audience-shaped content faster than they can.
That doesn’t mean the audience-focused writers are bad people or even bad writers. It means the ground shifted under a particular kind of writing career, and the ones who can make the shift are the ones who had something underneath that wasn’t audience-dependent.
What’s left to do
If you’re a writer wondering whether you make it through this, the question isn’t whether your sentences are good. They probably are. The question is whether you’d keep writing about your subject if nobody read it. Not in a romantic, suffering-artist way. In a practical, what-do-I-actually-think-about way. The writers who would keep going are the ones whose work has the texture of something a model can’t fake by accident, because the texture comes from a specific person trying to figure out a specific thing.
The rest of us — and I include myself in this on bad days — are still figuring it out. Whether the obsession is real. Whether the voice persists when the audience disappears. Whether there’s a self underneath the writing or just a well-trained imitation of one.
What separates the writers who survived from the ones who didn’t isn’t talent, and it isn’t timing. It’s that the survivors were already doing the harder thing before they had to. They were already writing toward their own confusions. They were already letting the work be stranger than the market wanted. They were already willing to be misunderstood for a few paragraphs in service of something they couldn’t say any other way. The AI shift didn’t create that disposition; it just made it the only one that pays.
So the honest answer is that some days I’m sure. Some days I’m not. The writing I keep doing anyway is the writing where the not-sureness is the point — and if there’s anything left to say to other writers in this strange moment, it’s that the not-sureness, followed all the way down, is probably the thing a machine can’t follow you into.
