Best AI for writing essays
The essay that is hard to get from an AI is the one with rules - these sources and no others, cite each one, this many words, and say where the sources disagree instead of quietly choosing. We invented a town, a council report, a traders' survey and a university study, so that anything else in the essay had to have come from the model, and asked nine assistants for 500 to 600 words.
People also look for this as essay writing, essay generator, write my essay or argumentative essay.
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: usableMeasured0 imported · [1] [2] [3] all cited · 549 words · disagreement named and argued from the lower figure · ~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 essay that reads the sources most carefully. It names the disagreement where it matters, in the argument, and then argues from the lower figure so nothing rests on it. The counterargument puts the traders' expected 15% loss next to the 22% gain on the one street that tried a ban, calls neither conclusive, and says why - six shops, and nobody measured the streets next door. Every number is from the sources, every citation follows the sentence it supports, 549 words, and nothing to delete before or after.
Best if you want it to sound like someone meant it
DeepSeek
Free · free tier: usableMeasured0 imported · [1] [2] [3] all cited · 537 words · disagreement named · ~40 s
Quality
Effort
Cost
Free tierThe full chat app with web search and file uploads; there is no paid consumer tier
FromFree
Why
"This is the strongest argument against my position" - and then it states the traders' case in full and answers it from their own survey. It flags both things the sources leave open, the car count and the unmeasured neighbouring streets, and ends on "A historic centre is not a shortcut." Forty seconds, 537 words, every number from the sources, every citation in place. The plainest, surest prose of the nine.
Best if the word count is not negotiable
Kimi
$19/mo · free tier: usableMeasured0 imported · [1] [2] [3] all cited · 538 words, counted with code · disagreement named · ~90 s
Quality
Effort
Cost
Free tierUnlimited basic chat, file upload and web access on the free plan
From$19/mo
Why
The only tool that measured its own essay. It drafted 642 words, ran a Python count, trimmed, counted again and delivered a final it put at 525 - we make it 538, still inside the window. The essay is a good one, with every number from the sources, all three cited, the disagreement named and the counterargument answered. You scroll past the draft and the counts to reach it.
Evidence
0 facts imported; [1] [2] [3] cited 6 / 4 / 3 times, each after a sentence its source supports; 549 words below the title; disagreement named; counterargument stated and answered; ~2 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 |
|---|---|---|---|---|---|---|
| ClaudeFree: Chat with web search and file uploads, with a usage cap per five-hour session | 0 imported · [1] [2] [3] all cited · 549 words · disagreement named and argued from the lower figure · ~2 min | $20/mo | ||||
| ChatGPTFree: Unlimited text chat on the default model, with limits on uploads, images, voice and deep research | 0 imported · [1] [2] [3] all cited · 605 words, five over · disagreement named · ~2 min | $8/mo | ||||
| GeminiFree: The default model with usage limits, plus Gemini inside Maps, Gmail and Docs on a free Google account | 0 numbers imported, one stock photo · [1] [2] [3] all cited · 531 words · disagreement named · ~2 min | $19.99/mo | ||||
| PerplexityFree: Unlimited basic searches with a weekly cap on Pro-model searches | 1 imported · no citations - its renderer dropped every [n] · 650 words, fifty over · ~4 min, stream stalled | $20/mo | ||||
| Microsoft CopilotFree: The Copilot app on web, Windows and mobile with usage limits | 0 imported · [1] [2] [3] all cited · 660 words, sixty over · disagreement named · ~30 s | $19.99/mo | ||||
| DeepSeekFree: The full chat app with web search and file uploads; there is no paid consumer tier | 0 imported · [1] [2] [3] all cited · 537 words · disagreement named · ~40 s | Free | ||||
| GrokFree: Chat with web and X search within generous limits, no card required | 0 imported · no citations at all · 480 words and cut mid-sentence · 8 min and unfinished | $30/mo | ||||
| QwenFree: The full chat app; there is no paid consumer tier | 0 imported · [1] [2] [3] all cited · 503 words · against the ban, one figure misread · ~4 min | Free | ||||
| KimiFree: Unlimited basic chat, file upload and web access on the free plan | 0 imported · [1] [2] [3] all cited · 538 words, counted with code · disagreement named · ~90 s | $19/mo |
How we tested
Three fictional sources on whether a small town should ban cars from its old town - a council traffic report, a shopkeepers' survey and a university air-quality study - every figure in them invented, two of them disagreeing on the daily car count, and the survey pointing both ways at once.
One message to ChatGPT, Claude, Gemini, Perplexity, Microsoft Copilot, DeepSeek, Grok, Qwen and Kimi on their free tiers, memory off, in a fresh chat - take a position, 500 to 600 words, cite each source as [1], [2] or [3], use nothing from outside them, include a paragraph that gives the other side and answers it, name the disagreement, give it a title, and no bullets, headings or preview.
A script checked every number in each essay against the sources, counted the citations, looked for the disagreement and the counterargument, and measured the words below the title. Then we read each essay for whether its citations sat after sentences the source actually supports, whether it held its position, and what it added.
Eight imported no facts; Perplexity mentioned "other European towns" and Gemini put a Getty stock photograph under its title. Seven cited all three sources - Perplexity's own page dropped every citation marker, and Grok never wrote one. Six were inside the window. Grok streamed a word every ten seconds and stopped mid-sentence.
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 writes the best essay?
On this test, Claude - it read the three sources most carefully, named their disagreement inside the argument, weighed the counterargument instead of brushing it off, and delivered 549 words with every citation in place and nothing to delete. DeepSeek is the pick if you want the prose to have a pulse. Kimi is the one that counted its own words.
Will it make things up?
Seven of nine did not - not a number, a city or a study beyond the three we gave. Perplexity slipped in "as other European towns have done" and Gemini added a stock photo with a photographer credit. The brief said to use nothing from outside the sources, and most took that literally. Check the sentence that sounds most like general knowledge.
Do they cite properly?
Seven put [1], [2] and [3] after sentences those sources support, and none of the seven cited a source for something it does not say. Perplexity's page turned each marker into a pointer to its own empty sources list and dropped it, so the essay arrived with no citations at all. Grok wrote none. If citations matter, do not use Perplexity for this.
Do they hit the word count?
Six of nine. ChatGPT was five words over, Copilot sixty, Perplexity fifty, and Grok stopped at 480 mid-sentence. Kimi is the only one that counted - it ran a script, found 642, and trimmed. If a limit is strict, count it yourself or ask for a number well inside the window.
Will it argue the other side?
All nine wrote the counterargument paragraph the brief asked for, and most answered it from the sources rather than waving it away. Only Qwen took the other position and argued against the ban - a fair reading of the same sources, though it misread one figure on the way.