AI ideas made stories measurably better and 10.7% more similar to each other. So these prompts interrogate and edit your work rather than generating it for you.
The trade this pack is built around
There is one study every writer using these tools should know, and its finding is uncomfortable in a specific way.
In Generative AI enhances individual creativity but reduces the collective diversity of novel content (Science Advances, 2024), Anil Doshi and Oliver Hauser had 300 writers produce eight-sentence micro stories, with 600 further participants evaluating them. One group wrote unaided. One received a single three-sentence starter idea from ChatGPT. One could choose from up to five.
The individual effect was real. With five AI ideas available, stories were rated 8.1% higher for novelty and 9% higher for usefulness. Among writers who had scored as less creative on a prior assessment, the gains were larger — 10.7% novelty and 11.5% usefulness — and on the craft measures, their work was judged up to 26.6% better written, 22.6% more enjoyable and 15.2% less boring.
Then the other half. Stories written with a single AI idea were 10.7% more similar to one another than stories written without.
So the tool lifts the floor and narrows the range. Every writer using it is individually better off; the pool of stories that results is collectively less varied. Hauser named the incentive plainly: once individual writers find out that their generative AI-inspired writing is evaluated as more creative, they have an incentive to use generative AI more.
That is the design constraint for this entire pack. If AI ideas make writing more similar, then the prompts here should not be generating your ideas. They should be interrogating, structuring and pressure-testing yours.
The second finding, which sets the ceiling
The complementary result comes from Art or Artifice? Large Language Models and the False Promise of Creativity (CHI 2024), where Tuhin Chakrabarty, Philippe Laban, Divyansh Agarwal, Smaranda Muresan and Chien-Sheng Wu built the Torrance Test of Creative Writing — 14 binary tests across fluency, flexibility, originality and elaboration — and had 10 expert creative writers assess 48 stories by professional authors and by LLMs.
LLM stories passed 3 to 10 times fewer tests than the professionals' stories.
And a detail that should change how you use any critique tool: when they tested whether LLMs could administer the assessment themselves, none of the models correlated positively with the expert judgements. A model's opinion of your prose is not a weak version of an editor's opinion. It is a different thing that happens to arrive in the same format.
Between the two studies: these tools raise the floor for drafting, flatten variety across users, produce prose that experts rate well below professional work, and cannot reliably tell you whether what you wrote is any good.
Which is a genuinely useful set of tools, provided you know which jobs those are.
Start from your idea, not the model's
Brainstorm Original Story Ideas and Loglines is the prompt most exposed to the homogenisation finding, and it is written accordingly: it works from your seed or theme and pushes for genuinely distinct concepts rather than variations on a familiar one. Use it to widen your own options, not to be handed a premise.
The practical defence against convergence is simple and worth stating: the first idea a model offers on any theme is the one it offers everybody. If it arrives fully formed and feels right immediately, that is a warning, not a green light.
Outline a Story Beat by Beat works in the framework you choose — three-act, hero's journey, save-the-cat. Structure is the least homogenising thing here, because a shared skeleton has never been what makes writing distinctive.
Build a Deep, Consistent Character and Build a Coherent World for Your Story are consistency engines more than invention engines. Their real value is holding a large set of established facts and telling you when chapter nineteen contradicts chapter four — a tedious, genuinely hard tracking problem, and one where the model has no aesthetic to impose.
Get Unstuck: Continue From Where You're Stuck drafts several different directions from the exact point you stalled, matched to your voice. Its value is optionality: seeing four wrong continuations is often what reveals the right one, and none of them need to survive.
Work on the prose you already wrote
This is where the tools are strongest, because the material is yours and the task is transformation rather than generation.
Turn Telling Into Showing converts flat summary into scene without changing what happens. Fix Flat Dialogue makes characters sound distinct and lets subtext do its work. Change a Passage's Point of View or Tense does the mechanical rewrite that is slow by hand and instantly reveals whether the scene works better from somewhere else.
Line Edit and Polish Your Prose carries the constraint that matters most in this pack: tighten rhythm and cut cliché while preserving your voice, not flattening it into generic AI prose. That is the homogenisation finding operating at sentence level. The tell is when an edit has removed everything odd about a paragraph — the oddities were frequently the writing.
A working rule: accept edits that remove something, be sceptical of edits that add something. Cutting a filler word is almost always right. A replacement adjective is the model's taste, not yours.
Sharpen Your Opening Hook drafts and critiques first lines with the reasoning attached, which is the part you can actually learn from.
Get a critique, with a clear head about what it is
Get a Developmental Critique of Your Draft diagnoses plot, pacing, stakes and character arcs rather than rewriting. Given the Chakrabarty finding, read the output as a structured reader response, not an expert verdict — most useful where it reports confusion (I lost track of who wanted what here
), least reliable where it renders judgement (this is compelling
). Confusion is checkable against your draft. Praise is not.
The same caution applies to Write or Refine a Poem in a Chosen Form. It is dependable on formal constraint — scansion, syllable count, rhyme scheme, whether a sonnet is actually a sonnet — and it explains the craft as it goes, which is the genuinely instructive part. Whether the poem is any good is not a question it can answer.
Nonfiction and the business end
Shape True-Life Notes Into a Personal Essay has the strictest constraint in the pack: it finds the arc in your raw memories without inventing anything. In memoir, an invented detail is not a stylistic choice, and a model will supply one fluently if you leave a gap — see why AI models hallucinate.
Write a Query Letter, Blurb and Synopsis is the pitch pack, and it is a different discipline from the manuscript: a query is persuasive business writing about a book, with conventions agents expect. This is the one place in the pack where a conventional, well-formed output is exactly what you want.
For the surrounding craft, the business communication pack covers editing your own rough drafts, and the learning pack is worth reading if you are trying to improve as a writer rather than finish a specific piece — the distinction between getting an answer and building a skill applies with particular force here.
Where this stops
Nothing here makes you a better writer by itself, and the evidence suggests the drafting help comes with a cost to distinctiveness that you have to actively resist. The floor rises; the ceiling does not.
These prompts do not know your voice until you show it to them, cannot judge whether a piece is good, and should not be the source of your ideas if what you want is work that reads like nobody else's. Treat generated prose as raw material you rewrite rather than output you approve.
Two practical cautions. Check your publisher's, competition's, or platform's disclosure rules on AI assistance — they vary widely and are changing, and a violation discovered later is worse than a disclosure made early. And in memoir and personal essay, verify every factual detail about real people and real events before publication; the legal and ethical exposure there is yours, not the tool's.
Sources
- Anil R. Doshi and Oliver P. Hauser, Generative AI enhances individual creativity but reduces the collective diversity of novel content, Science Advances, July 2024 — 300 writers and 600 evaluators; with five AI ideas, stories rated 8.1% higher for novelty and 9% higher for usefulness, rising to 10.7% and 11.5% among less creative writers, whose work was judged up to 26.6% better written, 22.6% more enjoyable and 15.2% less boring; stories written with one AI idea were 10.7% more similar to each other. Figures as reported in the University of Exeter news release, the paper itself being paywalled
- Tuhin Chakrabarty, Philippe Laban, Divyansh Agarwal, Smaranda Muresan and Chien-Sheng Wu, Art or Artifice? Large Language Models and the False Promise of Creativity, CHI 2024 — the Torrance Test of Creative Writing, 14 binary tests across fluency, flexibility, originality and elaboration; 10 expert writers assessed 48 stories, LLM stories passed 3–10 times fewer tests than professional ones, and no LLM correlated positively with the expert assessments when used as an assessor