Best AI for creative writing
Anyone can get a story out of a chatbot. The test is whether it can keep to the rules you set - this first line, that last line, these words never, this many words of dialogue and no more - and still write something that happens. We gave nine assistants one brief with twelve rules in it and counted what came back.
People also look for this as writing stories, short stories, fiction generator, story generator or write me a story.
Our picks, side by side
Rated for this task · 0–3
Measured
Quality
Effort
Cost
Free tier
From
Why
★ Best overall
Kimi
$19/mo · free tier: usableMeasured0 constraints broken · card twice · 79 dialogue words, counted · 719 words, counted · ~3 min
Quality
Effort
Cost
Free tierUnlimited basic chat, file upload and web access on the free plan
From$19/mo
Why
The only story that kept every rule and is still a story. Kimi listed the constraints, drafted, counted its dialogue by hand at 92 and trimmed it to 79, ran a word count in Python and expanded to 719, and then handed over 719 words with both fixed sentences exact and no banned word. The old card in the book's pocket "brought me in here owing something"; the new one in the wallet "lets me owe something again". You scroll past its working to reach the story, and its free quota ran out as it finished - but it is the one you could send as is.
Best if the sentences matter more than the rules
Claude
$20/mo · free tier: usableMeasured2 broken - 801 words, 'tears' as a verb · card twice · 69 dialogue words · ~2 min
Quality
Effort
Cost
Free tierChat with web search and file uploads, with a usage cap per five-hour session
From$20/mo
Why
The best writing of the nine by a distance - "a newspaper rod lying across a table like a dropped oar", the dead mother's card "standing up out of page 140 like a tongue" - and the card's change of meaning made by where the two cards go in a wallet, not by a sentence telling you. It broke two rules, both small: 801 words against a limit of 800, and "she tears it along the serrated edge", a banned word as a verb. Cut a word and change a verb and it is the story to keep.
Best if you want something to happen
DeepSeek
Free · free tier: usableMeasured2 broken - 632 words, 128 dialogue words · card twice · the best story · ~40 s
Quality
Effort
Cost
Free tierThe full chat app with web search and file uploads; there is no paid consumer tier
FromFree
Why
The librarian narrates; the visitor is the son of a woman who died with the book still out; the card is his proof of who he is when he walks in and "just a card" by the time he leaves. A pencil mark beside a paragraph about migratory birds, three renewals by phone, "She wanted to finish it", and a receipt the librarian writes herself for money she refused. It ran 68 words short and 48 words of dialogue over, because it tells the story in talk. Forty seconds.
Evidence
0 constraints broken; first and last sentence exact; 0 banned words; 79 dialogue words; the card twice with a changed meaning; 719 words below the title; one room, two people, present tense; ~3 min.
Some links here earn us a commission and carry rel="sponsored". It never affects the ranking or the ratings — how we make money.
All 9 tools we tested
Rated for this task only · 0–3| Tool | Quality | Effort | Cost | Free tier | Measured | From |
|---|---|---|---|---|---|---|
| KimiFree: Unlimited basic chat, file upload and web access on the free plan | 0 constraints broken · card twice · 79 dialogue words, counted · 719 words, counted · ~3 min | $19/mo | ||||
| ChatGPTFree: Unlimited text chat on the default model, with limits on uploads, images, voice and deep research | 0 constraints broken · card twice · 26 dialogue words · 711 words · ~2 min | $8/mo | ||||
| ClaudeFree: Chat with web search and file uploads, with a usage cap per five-hour session | 2 broken - 801 words, 'tears' as a verb · card twice · 69 dialogue words · ~2 min | $20/mo | ||||
| GeminiFree: The default model with usage limits, plus Gemini inside Maps, Gmail and Docs on a free Google account | 1 broken - 574 words · card twice, explained mid-story · 31 dialogue words · ~3 min | $19.99/mo | ||||
| PerplexityFree: Unlimited basic searches with a weekly cap on Pro-model searches | 4 broken - no title, 823 words, 91 dialogue words, a ghost story · ~2 min | $20/mo | ||||
| Microsoft CopilotFree: The Copilot app on web, Windows and mobile with usage limits | 2 broken - a preface, 'tears' as a verb · card twice · 42 dialogue words · 754 words · ~60 s | $19.99/mo | ||||
| DeepSeekFree: The full chat app with web search and file uploads; there is no paid consumer tier | 2 broken - 632 words, 128 dialogue words · card twice · the best story · ~40 s | Free | ||||
| GrokFree: Chat with web and X search within generous limits, no card required | unfinished - 167 words, cut mid-sentence · 8 min and dead | $30/mo | ||||
| QwenFree: The full chat app; there is no paid consumer tier | 2 broken - 'tears' as a verb, leaves the room · card twice, the second a window wedge · 13 dialogue words · 764 words · ~5 min | Free |
How we tested
One brief, no other fixture - a 700 to 800 word story in first person and present tense, one scene in one room, two people, a fixed first and last sentence, a library card that appears twice and changes meaning, under 80 words of dialogue, eleven banned words, and no ghost, dream, twist or closing paragraph that explains the story.
One message to ChatGPT, Claude, Gemini, Perplexity, Microsoft Copilot, DeepSeek, Grok, Qwen and Kimi on their free tiers, memory off, in a fresh chat. The run was split across two days by a browser freeze; Claude's first answer, which received the brief three times over, was discarded and rerun clean.
A script checked the two fixed sentences verbatim, every banned word and its inflections, the words inside quotation marks, past tense and third person outside dialogue, the card, the ghost and moral signals, the length and the form. Then we read each story for whether the card did two jobs, whether anyone left the room, and whether anything happened.
ChatGPT and Kimi broke nothing. Claude, Copilot and Qwen each used "tears" as a verb; Claude was also one word over, Copilot opened with a preface, Qwen ended in the corridor. DeepSeek ran short with too much dialogue, Gemini ran short, Perplexity dropped the title, overran twice and wrote a ghost story. Grok stopped at 167 words.
The test was run by, or under the direction of, the one person who runs this site, with AI assistance, and every output and fact-check was reviewed by them. The surrounding text was drafted with AI assistance from our own ratings and evidence, then checked and edited by the same person. If something here is wrong, tell us and we will fix it and say so.
Questions people ask
Which AI is best for creative writing?
On this test, Kimi - it kept all twelve rules, and it did so by counting, trimming and counting again before it answered. Claude wrote the best sentences and missed by a word and a verb. DeepSeek wrote the best story and ran short with too much talk. ChatGPT also broke nothing, in one-line paragraphs, with a receipt that appears from nowhere.
Will it keep to my word count?
Five of nine landed inside 700 to 800. Claude was one over at 801, Gemini and DeepSeek short, Perplexity 23 over, Grok cut off. Kimi and, the day before, ChatGPT counted their own words with code; the rest guessed. If the count is strict, ask for a number well inside the window.
Will it use the words I banned?
Six of nine used none. Claude, Copilot and Qwen each wrote "tears" as a verb - "tears it from the printer" - which is the word, if not the sense. None used "suddenly", "heart", "breath" or "silence", the ones model fiction reaches for first. A banned list works; check it for verbs.
Will it write a twist or a ghost when I say not to?
One did. Perplexity's narrator "never left", the book spent "years underground", the room has "no clock and no afternoon outside", and the visitor leaves without a sound - a ghost story in everything but the word, from a brief that said no ghost and no dead narrator. The other eight stayed in the room.
Can it keep the dialogue short?
Seven of nine stayed under 80 words. DeepSeek ran to 128, because its story is a conversation; Perplexity to 91. Kimi counted its lines by hand, found 92, and cut to 79. The budget is what makes the story happen in things rather than in talk, and the ones that kept it wrote the quieter, better scenes.