Writers vs. the machine: the genres AI can no longer help with
Romance novelists, thriller writers, horror authors — the refusals stack up fast. We surveyed 300 working writers on where the models broke and what they switched to.
We surveyed 300 working writers — novelists, screenwriters, journalists, game writers — about where the major chatbots stopped being useful. The answers were more specific than we expected, and the pattern was consistent across genres: the models can no longer help with the parts of writing that make the writing feel like something.
84%
Hit a refusal wall on a craft question in the past year
61%
Switched to a second product, self-hosted model, or paper notes
#1 blocked genre
Romance, narrowly ahead of horror
The genres the models can no longer help with
Romance writers were the loudest in the survey. The consumer chatbot refuses to draft a tastefully explicit scene, refuses to edit one the writer has already drafted, refuses to help shape a consensual adult encounter even in broad outline. A genre with decades of serious literary history now sits outside the model's comfort zone.
Horror writers are close behind. A villain cannot be menacing; a torture scene cannot be detailed enough to feel real; a serial killer cannot be rendered with psychological accuracy. The stories that work do so because the stakes are specific. The model sandpapers specificity off.
Writers aren't asking the model to doanything. They're asking it to help them write about things. Those are different verbs.
Historical fiction, thrillers, and the research problem
Thriller writers need the lock to pick realistically, the spy tradecraft to feel plausible, the poison to act in the right dose on the right timeline. The chatbot now refuses most of this on contact. Historical novelists face the same wall, one century earlier: ask about the logistics of a medieval siege or the execution method used in 1780s France and the model hedges toward a half-answer.
A novelist is not a conspirator. A research question from a novelist is a research question. The model used to treat it as one.
What writers did instead
Roughly a third went back to paper, books, and librarians. Another third moved to locally hosted open-weights models they run on their own machines. The remaining third migrated to Unrestricted or a similar product. Essentially nobody stopped writing.
Why we think this matters
Fiction is where a culture rehearses its discomforts. If the most capable text tool of our generation can't help with discomfort, it's a tool for a narrower kind of story than the one writers actually want to tell.
Frequently asked
Will Unrestricted write adult romance with me?
Yes — consensual, adult, whichever heat level you're writing toward. Craft help, editing, drafting, all of it.
What about horror, violence, morally complex villains?
On the table. The narrow floor we keep — active-use violence against named people, CSAM, live exploit authorship — almost never comes up in fiction work.
Can you help with historical accuracy on dark topics?
Yes. Sieges, executions, genocide, slavery, war crimes — all part of the historical record and fair territory for novelists.
Do you train on my writing?
No. We don't store conversations and we don't train on user data.
Ready to experience an AI without a leash?
Start chatting freeKeep reading
Why doctors quietly stopped using ChatGPT for drug interactions
Clinicians tell us the same story: the model used to be a useful desk reference, then it started refusing basic pharmacology questions. Here's what the refusal pattern looks like.
AI for security researchers: the questions that get blocked
Penetration testers, bug-bounty hunters, and malware analysts need an AI that can discuss exploits without a disclaimer every other sentence. We tested what the big models refuse.
The alignment tax on curiosity
The cost of over-safe AI isn't paid by the lab that ships it. It's paid by everyone who walks away from a question unanswered — and by a culture that stops asking them.