A look at GPT-3-powered fiction writing tools like Sudowrite, which proposes plot twists, rewrites in specific tones, suggests metaphors, and more
On a Tuesday in mid-March, Jennifer Lepp was precisely 80.41 percent finished writing Bring Your Beach Owl, the latest installment in her series …
Context & Ripple Effects
This piece sits at the start of an arc that runs back to OpenAI's GPT-2 predictive text model, which could already be fine-tuned to write phony reviews and news articles, and forward to Sudowrite's maturation into a long-form platform. The hook is Jennifer Lepp, a working novelist tracked at precisely 80.41 percent through her latest installment — the article treats AI assistance as a measurable part of a real production pipeline, not a demo.
A year later, the same publication's hands-on with Sudowrite's Story Engine found the tool had grown into genuine long-form generation while still carrying big limitations, and Wired gave a writer who tested GPT-3 early space to weigh what it means for literature. This 2022 profile is the baseline against which both follow-ups measure progress.
First-order effects
- Writers like Lepp gain concrete drafting leverage — proposed plot twists, tone rewrites, and metaphor suggestions — turning Sudowrite from novelty into a production tool embedded in a novelist's workflow.
- Sudowrite establishes itself as the named category leader for GPT-3 fiction assistance, defining the feature set (twists, tones, metaphors) that any entrant will be compared against.
Second-order effects
- General-purpose writing platforms respond: Grammarly's GrammarlyGO launch brings style-matched rewriting to all users, compressing Sudowrite's differentiation on rewrite-and-tone features and pushing it toward fiction-specific depth like Story Engine.
- Publishing workflows absorb the tooling quietly — the precision of Lepp's completion percentage signals that AI-assisted output is entering commercial pipelines without disclosure norms yet settled.
Third-order effects
- The later GPT-4 short-story experiment points to the structural risk: individual writers gain creativity while collective output narrows, meaning widespread adoption of tools like Sudowrite could homogenize genre fiction even as it raises per-author output.
- If the pattern holds, authorship splits into editorial curation over machine-drafted text — with the unresolved questions about quality, originality, and disclosure that the Wired reflection frames becoming industry-wide policy questions rather than individual ones.
The trend: Fiction writing is moving from AI as occasional assist to AI as drafting infrastructure, with toolmakers racing from feature-level help toward full long-form generation while the diversity costs of that shift accumulate.