I’ve been experimenting with Sudowrite’s “Show Not Tell” feature over the past several weeks, primarily for editing fiction drafts, and I’ve encountered a persistent limitation that I’m hoping the community can help me unpack. The tool performs admirably when I feed it a single, discrete sentence for transformation; it reliably generates more evocative, sensory-rich alternatives that align with the principle of showing rather than telling. However, my attempts to apply the feature at a paragraph level—which is often the more practical unit for revision—have yielded consistently unsatisfactory results.
When I submit an entire paragraph, the output typically manifests in one of two problematic ways. Either it rewrites only the first sentence or two, leaving the rest of the paragraph untouched and creating a jarring stylistic disconnect, or it attempts to rewrite the entire block but does so by treating each sentence in isolation, resulting in a collection of individually “shown” sentences that lack narrative cohesion and flow. The connective tissue of the paragraph, the rhythmic buildup and release of detail, seems to be lost in the process. This suggests the underlying model may be chunking the input by sentence and processing each sequentially without a holistic view of the paragraph’s intent and structure.
My working hypothesis is that this might be a constraint of the current implementation, perhaps related to token limits or a deliberate design choice to prioritize sentence-level operations for higher reliability. I’ve tried varying the paragraph length and complexity, and I’ve experimented with different directive phrasings in the command box, but the core issue remains. I am curious if others have developed effective workflows to circumvent this. For instance, is there a particular method of paragraph segmentation or a specific type of preparatory editing that makes the feature more effective on larger text blocks?
Furthermore, I’m interested in the broader discussion of how AI writing assistants can be architected to better understand and manipulate narrative units beyond the sentence. The paragraph is a fundamental building block of prose, governing pacing, focus, and thematic development. A tool that can intelligently suggest holistic “show don’t tell” revisions at this level would represent a significant advancement. For now, I am resorting to a manual, sentence-by-sentence application, which, while useful, is considerably more time-consuming and disrupts the editorial flow. Any insights, shared experiences, or technical speculations would be most welcome.
Let's keep it constructive