Writer and author — tasks, one by one
The unit of analysis is the task, not the job title. Each one below carries its direction, whether the judgement rests on evidence or on platform inference, the reasoning, and what it does not establish.
Every task on this page#
Writing fiction and books
Being augmented≈ Platform inferenceDrafting novels, stories and non-fiction books.
Prompted models still fall short, but the gap closes with training on an author's work. Stories generated by language models passed 3–10 times fewer creativity tests than stories by professional authors, and critics rated a world-class novelist above GPT-4 Turbo on every dimension. Yet in a preregistered study, expert readers who disfavoured prompted AI imitations of 50 authors' styles favoured them once the model was fine-tuned on each author's complete works, and detectors flagged only 3% of those outputs.
Studies of short excerpts and stories, not novels; they measure quality judgements, not what readers buy or writers earn.
Screenwriting and rewrites
Still human-led✓ Evidence-backedWriting and rewriting scripts for film and television.
The US screenwriters' contract keeps the writer human. It says neither traditional AI nor generative AI is a writer, that AI-produced material is not literary material or source material for credit, and that a company may not require a writer to use ChatGPT to write literary material; a writer needs the company's consent to use generative AI. The 2026 agreement preserves these terms to 2030.
One US collective agreement for signatory companies; non-union and other countries' work is not covered.
Short fiction and open submission markets
Automating✓ Evidence-backedSelling stories and books through open submissions and self-publishing.
Open markets are where machine-written volume lands. The editor of Clarkesworld, a science-fiction magazine, reported a surge of spam submissions in February 2023 and temporarily closed submissions after more than 50 arrived in one morning; the magazine does not consider AI-written or AI-assisted work.
One magazine's account from 2023; it shows volume arriving, not how many writers lost sales.
Rights and licensing
Still human-led✓ Evidence-backedControlling and licensing the use of one's books and scripts, including for AI training.
Courts have set some terms. In Bartz v. Anthropic, a federal judge held training on the books at issue was fair use but that using pirated copies for a central library was not; the case settled for $1.5 billion, about $3,000 per work. In Kadrey v. Meta, the judge ruled for Meta on that record but said the ruling does not mean such training is lawful and that market dilution was the plaintiffs' potentially winning argument. The 2026 screenwriters' agreement requires notice to the union when scripts are licensed to train AI.
Two district-court rulings that bind no other court, one settlement covering past conduct, and one US contract; the law is still developing.