TUE, OCTOBER 06, 2026
Independent · In‑Depth · Practitioner‑Tested
✎ General

Story Prompts Are for Page One. These Are for Page Ninety.

Eight prompts that use AI for what it is genuinely good at in fiction - interrogating a character you cannot articulate, mapping where a scene loses the reader, and naming the exact line an editor would stop at - rather than for generating prose you would only have to strip back out.

By AIToolsRecap October 6, 2026 8 min read 14 views
Home › Articles › General › AI Prompts for Creative Writing: 8 for When You...

Two Different Things Are Called a Writing Prompt

A story prompt is an idea. "A letter arrives forty years late." You have not started yet, and you want somewhere to begin. There are thousands of these and some excellent collections — Reedsy alone has over 2,700.

A craft prompt is for page ninety. You have a draft, chapter eleven drags, and you cannot work out why. You do not need an idea. You need a reader who will tell you the truth and is available at 1am.

This page is the second kind. If you want the first, go and get a premise somewhere that specialises in them — then come back when you are stuck.

Prompt 1 — Interrogate the Character

Here is a scene featuring my character: [paste].

Act as a sceptical reader. Ask me twenty questions about this character that the scene does not answer — what they want in this moment, what they are avoiding, what they would never say out loud, what they would do if the other character left the room.

Do not answer any of them. Do not suggest what the answers should be. Ask only.

Why it works: the instruction not to answer is the whole prompt. Let a model answer and it invents a generic character; make it ask and it finds the holes you were writing around. Three of the twenty will be ones you cannot answer, and that is the diagnosis.

Prompt 2 — Find Where It Drags

Here is a scene: [paste].

Go through it paragraph by paragraph and mark where a reader's attention would drop. Be specific — name the paragraph and say whether it drops because nothing changes, because something is explained that was already shown, because the stakes are unclear, or because the prose has gone abstract.

Do not rewrite anything. Do not be encouraging. I want the map, not the fix.

Why it works: you already know it drags. You cannot see where, because you know what you meant. A model has no memory of your intention, which for this one job makes it the better reader.

Prompt 3 — Fix Flat Dialogue

Here is dialogue from my draft: [paste].

For each line, tell me: what the character wants from the other person in that line, and what they are concealing. If a line has neither, say so — it is doing no work.

Then mark every line where both characters could be swapped without anyone noticing. That is the real problem.

Why it works: the swap test is brutal and it is the fastest diagnostic for dialogue there is. If your two characters can trade lines without the scene changing, they do not have distinct voices yet, and no amount of dialect or verbal tics will fix that.

Prompt 4 — Break the Middle

Here is my outline or synopsis: [paste].

The middle section sags. Tell me why, in structural terms: is the protagonist reacting rather than choosing, has the central question been answered too early, are there consequences that never arrive, is the opposition passive?

Then give me three ways to break it — not fixes, breaks. Things that would make the situation worse for my protagonist in a way the story has already set up.

Why it works: most sagging middles are a protagonist who has stopped making decisions. Asking for ways to make it worse is more productive than asking for solutions, because a story gets interesting when the character has to choose under pressure.

Prompt 5 — Show What I Summarised

Find every place in this passage where I have told the reader about something instead of putting them in it: [paste].

For each, quote the summary line and tell me what scene it is standing in for — who would be present, what would be happening, what the reader would see.

Do not write the scene. Describe what I would have to write.

Why it works: "show don't tell" is useless as advice because the problem is spotting it. This turns it into a list of specific lines with specific replacements, and leaves the writing to you.

Prompt 6 — Raise the Stakes Honestly

Here is my premise and current stakes: [paste].

Tell me what my protagonist actually stands to lose, and whether the reader has been given a reason to care about it yet. If the stakes are external — death, the world ending — tell me what the internal stake is, and whether I have established it.

If I have not, say so plainly. Do not invent one for me.

Why it works: external stakes are easy and inert. A reader cares when something the character wants is at risk, and that has to be established before the threat arrives. This finds out whether it was.

Prompt 7 — Generate a Premise in My Genre

I write [genre]. Here are three books in the space my work sits in: [titles]. Here is a sample of my prose: [paste 200 words].

Give me fifteen premises that would suit this voice specifically — not generic prompts for the genre, but ones that play to what my sample shows I am good at.

For each, say in one line what the central tension is. Discard anything that could appear on a generic prompt list.

Why it works: this is the one prompt here that generates rather than diagnoses, and the sample is why it beats a prompt list. A premise matched to your voice is worth fifty that are not.

Prompt 8 — The Honest Read

Read this as an acquiring editor with a full list who will reject most of what they see: [paste first 1,000 words].

Tell me the exact line where you would stop reading, and why. Then tell me what is working, specifically and briefly.

Do not soften it and do not lead with encouragement. I want the stopping point first.

Why it works: the stopping point is the single most useful note a draft can get. Writers rarely ask for it because it is uncomfortable, and friends will not give it.

What AI Is Actually Bad At Here

Every page about AI and fiction oversells this, so plainly:

AI is poor at prose you would want to keep. Ask it to write your scene and you get competent, rhythmically uniform, slightly generic text that sounds like everyone. You will spend longer removing its voice than you would have spent writing the scene.

It is poor at knowing what is interesting. It will call your work engaging and compelling unless you explicitly instruct it not to — which is why half the prompts above contain an instruction about tone.

It cannot hold your whole book. It will contradict chapter three while fixing chapter eleven, and do it confidently.

What it is genuinely good at is interrogation. Asking questions, finding structural holes, noticing what you summarised, spotting where two characters sound identical. Those are jobs where having no stake and no memory of your intentions is an advantage rather than a limitation.

Use it as a reader who is always available and never kind. Not as a writer.

FAQ

What are the best AI prompts for creative writing?

The ones that ask rather than answer. Prompts that make a model interrogate your character, map where a scene drags, or name the line an editor would stop at are far more useful than prompts asking it to write. Generation is where AI is weakest; diagnosis is where it is strongest.

Are these creative writing prompts for authors different from story prompts?

Yes. A story prompt is a premise for before you start. These are craft prompts for when you have a draft and something is not working. Different problem, different tool.

Can ChatGPT write fiction for me?

It can produce fiction. It will read as competent and rhythmically uniform, and you will spend longer stripping its voice out than you would have spent writing. Use it on the draft you wrote, not instead of writing one.

Do these work on Claude, Gemini and Grok?

Yes. They are structural instructions rather than model tricks. Longer passages benefit from a larger context window, so a model with a million-token context handles whole chapters better.

What should I never ask AI to do with my fiction?

Judge whether it is good. It will say yes. Ask it where a reader would stop instead — that question has an answer it can actually find.

How do I stop it being encouraging?

Instruct it explicitly. "Do not soften it", "do not lead with encouragement", "I want the stopping point first". Without that, default politeness makes the feedback worthless.

These eight are a starting set. Our library has 744 prompts across 123 collections for ChatGPT, Claude, Gemini, Grok and Kimi — the ChatGPT writing collection and the Claude writing collection are the nearest neighbours to this page.

Tags
Best AI ToolsAI GuideChatGPTProductivity2026
⚑

Spot an inaccuracy?

We verify facts before publishing and correct errors promptly. If something in this article is wrong or outdated, let us know.

Report an error →
💡 AI Tools prompts
Prompt Guide
Best Claude AI Prompts for SEO (2026) — Content, Technical, and Comparison SEO
Claude Sonnet 5 and Opus 5 are strong for SEO work that requires writing quality, structured analysis, and long-form content generation. With 1M context, Claude can analyse an entire site's content structure, compare competing pages, and write complete article drafts in one session. These prompts cover the full SEO workflow: keyword research synthesis, content briefs, on-page optimisation, meta descriptions, technical audit interpretation, and comparison content that ranks above AI Overviews.
Get Prompts →
Prompt Guide
Best ChatGPT Prompts for SEO (2026) — GPT-5.6 and Browse
ChatGPT with GPT-5.6 Sol and Browse enabled is a capable SEO research tool — it can search the live web, analyse SERP results, and synthesise content briefs in a single session. GPT-5.6 Terra at $2.50/M offers a cost-efficient option for high-volume SEO content generation. These prompts are optimised for ChatGPT Plus with Browse, the ChatGPT Work product for larger projects, and the OpenAI API with web_search tool enabled.
Get Prompts →
Prompt Guide
Best Claude Opus 5 and Sonnet 5 Prompts for Writing (2026)
Claude Opus 5 and Sonnet 5 consistently produce the highest-quality long-form writing of any AI model in July 2026 — a lead documented across writing benchmarks and user testing since Claude 3 Opus. With 1M context and 128K output on Opus 5, Claude can write book chapters, complete reports, and long-form content without truncating. Sonnet 5 at $2/$10/M (intro through August 31) is the best value writing model available. These prompts are optimised for claude.ai Pro/Max, Claude Cowork, and the API.
Get Prompts →