Best AI for writing
We ran five writing tasks - a short story, a piece of fan fiction, a worldbuilding entry, a chapter from a story bible and an argued essay - as one brief each to nine assistants on their free tiers, and checked every constraint. Claude was picked on all five.
People also look for this as best ai for writers, best ai writer or best ai to write with.
Our picks, side by side
Rated for this task · 0–3
Measured
Quality
Effort
Cost
Free tier
From
Why
★ Best overall
Claude
$20/mo · free tier: usableMeasuredpicked on all 5 · overall on 3
Quality
Effort
Cost
Free tierChat with web search and file uploads, with a usage cap per five-hour session
From$20/mo
Why
It reads the source more carefully than anything else we put these briefs through. Given a story bible it used the details the traps were made of - a stair rail "built with her in mind", a photograph face down in a drawer. Given Pride and Prejudice it caught an error in our own brief, said so in a note, and wrote the scene the book's own chronology allows. Given three sources that disagreed it named the disagreement inside the argument and then argued from the lower figure, so nothing rested on it. Given six invented facts about a town it made them into a place rather than restating them. The cost is length: it ran over the window on two of the five, and on the short story it broke two constraints to write the best prose in the run.
Best for voice
DeepSeek
Free · free tier: usableMeasured4 picks in 5 tasks · 0 overall
Quality
Effort
Cost
Free tierThe full chat app with web search and file uploads; there is no paid consumer tier
FromFree
Why
Never the overall pick on any of the five, and a named pick on four - always for the same thing. It writes the line that sounds like a person said it: "a question wearing the dress of a statement", a character who asks not to be made a fool of because "I do not think I could bear gentle". Its timelines explain the facts instead of restating them. Take it when the sentences matter more than the brief, because it is the one that ignores a length window - 45 words over on one task and 202 over on another.
Best when the rules are strict
Kimi
$19/mo · free tier: usableMeasured3 picks · overall on 1 · 2 no-answers
Quality
Effort
Cost
Free tierUnlimited basic chat, file upload and web access on the free plan
From$19/mo
Why
The only tool that measures its own work. On the short story it listed the constraints, counted its dialogue by hand, ran a word count in Python and trimmed to fit - the one entry that kept every rule and was still a story. On the essay it drafted, counted, trimmed and counted again. The catch is availability: its free quota ran out partway through our Writing run and it returned nothing at all on two of the five tasks.
Best for a clean first draft
ChatGPT
$8/mo · free tier: usableMeasured2 picks in 5 tasks · 0 overall
Quality
Effort
Cost
Free tierUnlimited text chat on the default model, with limits on uploads, images, voice and deep research
From$8/mo
Why
The safest of the nine and the plainest. It hit the window on the tasks where the others drifted, cited every source, bent no facts, and handed the chapter over as a document you can copy whole. What you get back is correct and a little flat - facts restated rather than used, and a page of one-line paragraphs - so it is a first draft to edit rather than a piece to publish.
Evidence
Overall pick for storytelling (0 facts contradicted, every window kept), writing a book (0 facts broken, the bible's small facts used, 26 words over) and writing essays (0 imported facts, all three sources cited, 549 words); best-for-canon on fanfiction, where it was the only tool to catch the brief's own timing error; best-for-prose on creative writing, where it broke two constraints.
Some links here earn us a commission and carry rel="sponsored". It never affects the ranking or the ratings — how we make money.
Every writing task
5 testedHow we tested
No separate test for this page. Each of the five Writing tasks below was its own run: one brief, sent once to nine tools on their free tiers in fresh chats with memory off, on the same day. Every answer was checked by a script committed with it - word counts, section counts, banned names, citations - and then read line by line against the source it was given.
This page counts where each tool was picked across those five runs and reads why. The ratings are the median of each tool's ratings across the five tasks, so a tool that is excellent once and absent twice does not come out ahead of one that was solid five times.
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
Is Claude the best for every kind of writing?
No. It was the overall pick on three of the five and picked on all five, but on the short story it broke two of the brief's constraints to write the best prose in the run, and it went over the length window on two tasks. If the brief is strict, Kimi kept every rule; if the sentences matter most, DeepSeek wrote the best of them.
Why is DeepSeek never the overall pick?
Because it does not respect a length window - 45 words over on one task, 202 on another - and the overall pick has to do what was asked. It was a named pick on four of the five, every time for voice, feeling or the way it turned facts into history.
What happened to Kimi?
Its free quota ran out partway through the run and it returned nothing on two of the five tasks. Where it did answer it was excellent, and it is the only tool that counts its own words with code rather than guessing. The quota resets, but it is a real limit on a free plan.
Which tools were not picked at all?
Gemini, Qwen and Perplexity were picked on none of the five, and Grok on none - its answers repeatedly died mid-stream, once after 454 words with the entry unfinished. They are rated in the table for each task so you can see how they scored.