55 Creative Writing Prompts to Use With AI

The risk of writing with AI is not that it writes badly. It is that it writes competently and characterlessly, and that competent characterless prose is very easy to accept when you are tired and the deadline is close.

So these fifty-five are a different kind of prompt from the ones you will find elsewhere. They do not give you a premise to write about. They give you ways to put a model to work: interrogating you, generating options you then choose between, and telling you what is wrong with what you have already written. The writing stays yours.

If you want premises rather than method, our writing prompts collection is organised by how much time you have, narrative writing prompts by craft element, and short story prompts by story engine. This page is about the working relationship rather than the idea.

• Never paste a generated sentence into your draft. Read it, decide what it got right, close the window and write your own version. The moment you start copying, the voice starts drifting and you will not notice until somebody who knows your work says something.
• Ask for questions, not answers. A model that interrogates you produces material only you have. A model that answers produces material anybody could have got. Most of the prompts below are built on this one distinction.
• Ask what the obvious version is, then avoid it. A model is an excellent map of the predictable, which makes it an excellent guide to what not to write. This is the single most useful thing it does for fiction.
• Use it after the draft, not before. Anything you ask before you have written tends to narrow what you write. Anything you ask after tends to sharpen it. If you only change one habit, change the order.

A request that will write your story for you:

Write a scene where two old friends meet after ten years apart.

The same subject, put to work:

I am writing a scene where two old friends meet after ten years apart. Do not write it. Instead, tell me the six things most writers would put in this scene, so I can avoid all of them. Then ask me four questions about these two specific people that would make the scene impossible for anybody else to have written.

The second one costs four minutes and leaves you with something to write. The first leaves you with something to edit, which is a worse position disguised as a better one.

Eleven prompts for the blank page. None of them writes anything for you, because the problem at this stage is almost never a shortage of words.

1. Map the obvious first

I want to write about [subject or situation]. Before I start, tell me the eight most predictable ways this gets written: the usual opening, the usual structure, the usual images, the usual ending. I want a list of things to avoid, not suggestions.

2. Interview me until something appears

I know roughly what I want to write about but cannot start: [describe the rough area]. Interview me. One question at a time, follow up on the vaguest part of every answer, and do not offer any material of your own. After twelve questions, give me back only the specific things I said.

3. Find the real subject

Here are three things I keep wanting to write about: [list them]. Tell me what they have in common that I might not have noticed. If they have nothing in common, say so rather than forcing a connection.

4. The wrong entry point

Here is where I was planning to start this piece: [describe]. Suggest five other places it could start instead, including one much later than I intended and one that begins after the interesting part is over. For each, tell me what it forces me to do differently.

5. Constraints instead of ideas

Do not give me an idea. Give me six constraints for a piece of writing - a length, a prohibition, a required element, a point of view restriction. Make at least two of them uncomfortable. I will supply the subject myself.

6. What I am avoiding

Here is what I have been circling for weeks without writing: [describe]. Ask me the question I am avoiding about it. Ask it plainly and do not soften it, then wait for my answer before saying anything else.

7. Ten first lines, none of them mine

Write ten opening lines for a piece about [subject], each using a different structure, and label the structure. I am not going to use any of them. I want to see the range so I can find the one you did not think of.

8. Argue me out of it

Here is the piece I am planning to write: [describe]. Tell me honestly whether it is worth writing, what has already been written that covers this ground, and what would have to be true about my version for it to justify the reader's time.

9. Scale it down

I want to write [describe the large thing] and it is too big to start. Break it into the smallest piece that would still be complete on its own and worth reading. Tell me what that piece is, what it leaves out, and whether the rest is needed at all.

10. Fifteen minutes, no thinking

Give me one instruction I can follow for fifteen minutes without deciding anything. It should be specific enough that I never have to stop and think about what to do next, and open enough that I could not predict what I will have written.

11. What only I could write

Here is what I know, where I have worked, and what I have seen that most people have not: [describe]. Ask me six questions designed to find the piece that only I am in a position to write. Do not propose subjects yourself.

Expand

Prompt 1 is the one to start with every time. Knowing the predictable version is worth more than any idea a model could hand you, because it tells you where not to go.

Eleven prompts for people on the page. A model asked to invent a character produces a plausible average of every character it has read. A model asked to interrogate yours produces something only you could have written.

12. Interview my character

Here is a character I am writing: [describe them]. Interview me as though you were trying to work out whether I actually know this person. Ask about ordinary things - what they eat, what they lie about, what they are bad at - and follow up whenever I answer with something generic.

13. The contradiction

Here is my character: [describe]. Identify the contradiction in them, the place where two things I have said do not fit together. If there is no contradiction, tell me so, because that is the problem - a character without one reads as an argument rather than a person.

14. What they want versus what they need

Here is my character and their situation: [describe]. Tell me what they say they want, what they actually want, and what they would need for the story to be worth telling. Then tell me which of those three I have not made clear on the page.

15. Test the dialogue

Here is a page of dialogue between two characters: [paste it]. Remove the speech tags and tell me whether you can still tell who is speaking. Where you cannot, tell me which line gave it away and what would distinguish them - vocabulary, sentence length, what they refuse to say.

16. The thing they will not say

My character is in this situation: [describe]. Tell me the one thing they would never say out loud here, and three things they might say instead of it. I want the evasions, not the confession.

17. Read them as a stranger

Here is the first appearance of my character on the page: [paste it]. Tell me what a reader knows about them after this passage, what they assume, and what they get wrong. Do not tell me what I intended - tell me what is actually there.

18. The minor character who is doing nothing

Here is my cast: [list them with their function]. Identify anybody who exists only to receive information or move a scene along. For each, tell me whether to give them something of their own or cut them and redistribute what they do.

19. Give the antagonist a case

Here is the character standing in my protagonist's way: [describe]. Make the strongest case that they are right. Not a sympathetic backstory - an actual argument that a reasonable person would find hard to answer. Then tell me whether my protagonist has a response.

20. My own voice, described

Here are three pieces I have written: [paste them]. Describe my voice technically - sentence length and variation, how much I explain, what I do with dialogue, my habits and my tics. Then tell me which habit is a strength and which one I lean on when I am not sure what to write.

21. Where the voice slipped

Here is a passage where I think the writing stopped sounding like me: [paste it, with a passage that does sound like me for comparison]. Identify exactly where it changes and what changed - rhythm, vocabulary, distance from the subject. Do not rewrite it.

22. The character I keep writing

Here are the main characters from four things I have written: [describe them]. Tell me what they have in common that I have not noticed. Be blunt about it, including if the answer is that I have written the same person four times.

Expand

Prompt 15 is the fastest diagnostic in this whole article. If the dialogue does not survive having its tags removed, no amount of description will fix the characters underneath it.

Eleven prompts for structure and setting. The useful move here is asking a model to find the hole rather than to fill it, because a hole you found is a decision and a hole it filled is a compromise.

23. Find the hole

Here is my plot so far: [describe it]. Do not suggest solutions. Tell me where it does not hold together: what a reader would question, what happens too conveniently, and which character does something they would not do. List them in order of how badly they damage the story.

24. The scene that is not earning its place

Here is my scene list with a one-line summary of each: [paste]. Identify any scene where nothing changes between the start and the end. For each, tell me whether to cut it, merge it, or give it something to do.

25. Three rules, and what breaks them

Here are the rules of the world I am building: [state them]. Find the three situations where these rules produce a result I probably have not thought about, and the one place where two rules contradict each other. Do not fix anything - I want to know what I have made.

26. What the setting does to people

Here is the place my story is set: [describe]. Tell me what living there would do to somebody over twenty years - what it would make them good at, what it would make them frightened of, and what they would consider normal that a reader would not.

27. Reorder it

Here are my scenes in their current order: [list]. Propose three different orders, including one that starts much later and one that withholds something I currently reveal early. For each, tell me what the reader knows at the halfway point and how that changes the ending.

28. The ending I am heading towards

Here is my story and the ending I have planned: [describe both]. Tell me whether the ending is the one the story has been promising or the one I wanted. If they differ, tell me what I would have to change earlier to earn the ending I want.

29. Off-stage instead

Here are the dramatic events in my story: [list them]. For each, tell me what would be gained by keeping it off the page and showing only the aftermath. Identify the one that would be strongest handled that way.

30. What the reader knows and when

Track what the reader knows at each point in my draft: [paste the outline]. Tell me where they know more than the protagonist, where they know less, and where I have withheld something for no reason other than to withhold it.

31. The research I actually need

My story involves [describe the specialised subject]. Tell me what I need to get right for it to feel authentic and what readers will not notice, and be honest about which parts of this you may be unreliable on so I know what to verify with a real source.

32. Same story, other genre

Here is my story: [describe]. Tell me how it would be structured if it were a thriller, a comedy and a literary short story. I am not changing genre - I want to see which structure it is closest to already, because that may not be the one I thought.

33. What it is about

Here is my draft: [paste or summarise]. Tell me what it is about, as distinct from what happens in it. Then tell me whether the thing it is about is present in the ending. Do not be generous.

Expand

Notice how many of these begin with do not suggest solutions. A model will fix your plot if you let it, and a plot fixed by a model is a plot that now belongs to nobody.

Eleven prompts for while you are actually writing. This is the dangerous stage, because it is when the temptation to paste is highest. Every prompt here is designed so that the output is not usable as text, only as information.

34. Options, not prose

I am stuck on how this scene continues: [describe the situation]. Give me eight things that could happen next, each in one line, none of them written out as prose. Include two that would be a mistake and say which they are.

35. The word I cannot find

I need a word that means roughly [describe the meaning] but with [describe the connotation you want and the one you want to avoid]. Give me twelve candidates with a note on the register of each. Do not write the sentence for me.

36. Where is this going wrong

Here is a paragraph I have rewritten five times and still do not like: [paste it]. Tell me what the actual problem is - is it the information, the order, the rhythm, or that it should not exist? Diagnose it. Do not rewrite it.

37. The research question, mid-sentence

I need one factual detail to keep writing: [state the question]. Answer briefly, and tell me how confident you are and whether this is something I should verify before publishing. Then stop - I do not want the surrounding paragraph.

38. Read it back to me

Here is what I have written: [paste]. Tell me what you understood, in your own words and in the same order. Where your version differs from what I meant, the fault is mine and I want to know exactly which sentence caused it.

39. The sentence doing too much

Here is a passage: [paste]. Identify any sentence that is carrying more than one idea and any that is carrying none. Mark them. Do not fix them - I want to decide which ones were deliberate.

40. Am I explaining too much

Here is a passage: [paste]. Point out every place where I have told the reader something they could have worked out from what is already on the page. For each, tell me what would be lost by cutting it, so I can judge whether the redundancy was doing work.

41. Keep me honest about pace

Here is a chapter: [paste]. Map the pace - where it slows, where it rushes, where a reader would put the book down. Do not suggest changes. Just tell me where the attention drops and what is happening at that point.

42. The next line I would write

Here is my draft so far: [paste]. Tell me what the predictable next sentence would be, given everything before it. I want to know so that I can write a different one.

43. Stuck in the middle

I am two-thirds through and have lost interest in my own draft: [describe where you are]. Ask me five questions to work out whether the problem is the story, the section I am in, or simply that I have been writing it too long. Do not reassure me.

44. Rebuild the momentum

Summarise back to me what I have written so far in this draft: [paste]. Just the events and the decisions, no interpretation. I have lost the thread and need to see the shape of it before I can start again.

Expand

Prompt 42 is the one that changes how the tool feels. Once you start using a model to find out what the obvious sentence is, it stops being something that writes for you and becomes something that tells you where everybody else went.

The last eleven, and where a model is least risky and most useful. You have already written the thing, so nothing here can replace your voice - it can only tell you what is on the page. Our AI writing tools handle the mechanical passes once the judgement calls are made.

45. What a first reader would say

Here is my finished draft: [paste]. Read it as somebody encountering it for the first time with no goodwill towards me. Tell me where you were confused, where you were bored, and where you stopped believing it. Be specific about the sentence in each case.

46. The three things I lean on

Here is my draft: [paste]. Identify the three constructions I use most often - a sentence shape, a transition, a way of ending paragraphs. Tell me how many times each appears. Do not tell me to vary them; I want to see the count and decide.

47. Cut it by a fifth

Here is a piece that is twenty per cent too long: [paste]. Tell me what to cut - by line reference, not by rewriting. List everything you would remove, and mark the one cut you are least comfortable with so I can overrule it.

48. Does the opening earn the rest

Here are my first two paragraphs and a summary of the whole piece: [paste both]. Tell me what the opening promises and whether the piece delivers it. If it does not, tell me whether to change the opening or the piece.

49. The ending, tested

Here is my final paragraph: [paste]. Tell me whether it resolves, explains, or simply stops. Then tell me what would be lost if I deleted the last sentence, and the last two.

50. Where I hedged

Here is my draft: [paste]. Find every place where I softened a claim, added a qualifier, or backed away from something I clearly meant. List them. Tell me which ones are honest caution and which are nerves.

51. Read it aloud for me

Here is a passage: [paste]. Identify every sentence that would be difficult to read aloud - too long to hold, awkward in the mouth, ambiguous in stress. Mark them without fixing them, and tell me which one is worst.

52. Does it sound generated

Here is my draft: [paste]. Tell me honestly which parts read as though a machine wrote them - the tidy summaries, the balanced sentences, the transitions that explain what just happened. Be blunt, including about passages I wrote entirely myself.

53. Mechanics only

Check only spelling, punctuation, consistency of names and dates, and tense agreement in the text below. Do not change any wording, do not suggest improvements, and do not comment on style.

54. Is it finished

Here is my draft and here is what I have already changed three times: [describe both]. Tell me whether further revision would improve it or whether I am now moving things around. Be direct - I would rather be told to stop than encouraged to continue.

55. What I learned writing it

Here is the finished piece and here is what I struggled with while writing it: [describe both]. Tell me what this reveals about my habits, and the one thing to watch for in the next piece. Do not praise the work.

Expand

Prompt 52 is worth running on everything, including work you wrote without any help at all. The habits that make writing sound generated - the tidy summary, the balanced sentence, the transition that explains itself - are human habits first, and a model is unusually good at spotting them.

Almost nothing on this page asks a model to write, which means model choice matters less here than in most guides. What does matter is whether it will follow a negative instruction - do not rewrite it, do not suggest solutions, do not be generous - for more than one exchange. That is worth testing directly: give the same diagnostic prompt to two models from the model list and see which one starts helpfully rewriting your paragraph anyway.

If you do want drafting help rather than diagnosis, an AI story generator is the honest place for it, and our fun writing prompts are built around games and constraints instead. Chat Smith is free to try.

And the test that matters more than any of the fifty-five: read the finished piece and find one sentence only you would have written. If you cannot, the draft went through too many hands, and one of them was not yours.