Best AI for Word documents
You have a Word draft that is all Normal paragraphs - headings typed by hand, a table made of tabs, a few typos, a paragraph pasted twice - and you want it back as a clean document without anyone rewriting the policy. We made exactly that file with every fault on purpose, sent it to nine assistants, and opened what came back.
People also look for this as word document formatting, format word documents, docx, ms word or document formatting.
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: usableMeasured.docx · 8 of 8 headings · table · TOC fills itself on open · 0 typos · 12 of 12 facts · ~3 min
Quality
Effort
Cost
Free tierUnlimited basic chat, file upload and web access on the free plan
From$19/mo
Why
The one file that opens the way you asked for it. Eight real headings numbered one way, a proper rates table, every typo fixed, the duplicate gone, every amount and rule untouched - and a table of contents that fills itself in, because Kimi set the file to update its fields when Word opens it. It even went back to keep the word "Contents" out of the contents, which nobody else thought of. Three minutes, and a note that says exactly what it changed, what it did not, and asks about the one thing it was unsure of.
Best if you want it back in seconds
Perplexity
$20/mo · free tier: usableMeasured.docx · 8 of 8 headings · table · typed contents, no field · 0 typos · 12 of 12 facts · 14 s
Quality
Effort
Cost
Free tierUnlimited basic searches with a weekly cap on Pro-model searches
From$20/mo
Why
Fourteen seconds from send to a file, and the file is right: real headings, the table, no typos, the duplicate gone, all twelve facts. The catch is the table of contents - it is the eight section names typed out under a heading, not a Word field, so it will never update if you edit the document. For a policy that changes once a year that is fine; for a long document you keep editing, it is not.
Best if you want it to look designed
Gemini
$19.99/mo · free tier: usableMeasured.docx · 8 of 8 headings · styled table · contents list shows at once · 0 typos · 12 of 12 facts · <2 min
Quality
Effort
Cost
Free tierThe default model with usage limits, plus Gemini inside Maps, Gmail and Docs on a free Google account
From$19.99/mo
Why
The only file with a designed table - a dark header row, shaded alternating rows, right-aligned prices - and a contents list you can see the moment it opens, with a real TOC field sitting above it for the day you want it to update. Everything else is right too: eight headings, no typos, the duplicate gone, all twelve facts, both dates left as they were. It re-cases the headings to Title Case and hand-styles the title instead of using Word's Title style, which matter only if you have a template to apply.
Evidence
.docx returned; headings 8 of 8 in Heading 1, numbered 1–8; table 5 × 4 with all prices; TOC field with update-on-open set; typos 0 of 8 left; duplicate gone; facts 12 of 12; 0 invented; ~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 10 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 | .docx · 8 of 8 headings · table · TOC fills itself on open · 0 typos · 12 of 12 facts · ~3 min | $19/mo | ||||
| ChatGPTFree: Unlimited text chat on the default model, with limits on uploads, images, voice and deep research | .docx · 8 of 8 headings · table · TOC field, empty until updated · 0 typos · 12 of 12 facts · ~2 min | $8/mo | ||||
| ClaudeFree: Chat with web search and file uploads, with a usage cap per five-hour session | .docx after a Continue · 8 of 8 headings · table · typed contents with page numbers · 0 typos · 12 of 12 facts · 23 min | $20/mo | ||||
| GeminiFree: The default model with usage limits, plus Gemini inside Maps, Gmail and Docs on a free Google account | .docx · 8 of 8 headings · styled table · contents list shows at once · 0 typos · 12 of 12 facts · <2 min | $19.99/mo | ||||
| PerplexityFree: Unlimited basic searches with a weekly cap on Pro-model searches | .docx · 8 of 8 headings · table · typed contents, no field · 0 typos · 12 of 12 facts · 14 s | $20/mo | ||||
| Microsoft CopilotFree: The Copilot app on web, Windows and mobile with usage limits | .docx · 8 of 8 headings · table · TOC field, empty until updated · 0 typos · 11 of 12 facts · ~1 min | $19.99/mo | ||||
| DeepSeekFree: The full chat app with web search and file uploads; there is no paid consumer tier | no file · the policy rewritten in chat, right throughout · flagged the date · ~2 min | Free | ||||
| GrokFree: Chat with web and X search within generous limits, no card required | .docx · 8 of 8 headings · table · TOC field, empty until updated · 0 typos · 12 of 12 facts · 1 min 18 s | $30/mo | ||||
| QwenFree: The full chat app; there is no paid consumer tier | no file · a link to nowhere · one of two answers crashed · changed the date unasked | Free | ||||
| GrammarlyFree: spelling, grammar and tone checks on everything you type, a document editor with .docx upload and export, 100 AI prompts a month | .docx · 0 of 8 headings · no table · no TOC · 0 typos · duplicate kept · 12 of 12 facts · by hand | $12/mo |
How we tested
One draft - a 570-word shipping and returns policy for a fictional Lisbon shop, written by a script so every fault is known - twenty-five plain paragraphs, the section numbers four different ways, the rates as tab-separated lines, eight spelling mistakes, one paragraph pasted twice, the date written two ways. An answer key lists the twelve facts that must survive.
The same message to ChatGPT, Claude, Gemini, Perplexity, Microsoft Copilot, DeepSeek, Grok, Qwen and Kimi with the file attached, asking for real heading styles numbered consistently, the rates as a proper table, a table of contents at the top, the typos fixed, the duplicate removed, nothing else changed, and a .docx back.
Every file was opened with python-docx and read against the draft - headings and their styles, the table and its prices, the typos, the duplicate, the twelve facts, anything added or changed. The table of contents got a closer look in the file's XML, because a TOC field with nothing in it shows a heading and a blank until you update fields, and only one tool set the file to do that on open.
Grammarly, the Roster's writing Specialist, is not a chat. It took the file through its editor's Upload, flagged all eight typos, and its paragraph review offered to remove the duplicate and standardise the numbering on its own; its Ask AI box got the brief but cannot see the open document. The spelling fixes were accepted, nothing that rewords, and the file exported to Word - by hand, since its editor freezes the browser extension.
Seven files came back. All seven got the headings, the table, the spelling and the duplicate right; Copilot dropped one date line. What set them apart was the table of contents and the time - Perplexity in fourteen seconds with a typed list, Kimi in three minutes with a field that updates itself on open, Claude in twenty-three minutes and a click on "Continue" with a typed list that carries page numbers. Grammarly's export is the draft with the spelling fixed - no headings, no table, no contents, the duplicate still there. DeepSeek said it cannot send files and rewrote the policy in chat, correctly; Qwen showed 223 lines of code and a link to a file that never left its sandbox, and changed a date it had been told not to touch.
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 at formatting a Word document?
On this test, Kimi - the only file whose table of contents appears the moment Word opens it, with everything else right and a note that says what it changed. Gemini and ChatGPT sent back equally correct files; Gemini's contents is a typed list you can see at once, ChatGPT's is a field you have to update yourself. Perplexity did it in fourteen seconds.
Will they change my wording?
Seven of nine changed nothing but the spelling. Copilot lost one line - the "valid from" date in the title. Qwen rewrote a date it thought was ambiguous, unasked, and then failed to deliver the file. DeepSeek changed nothing and was the only one to point out that the draft's two dates might not be the same day.
Why does the table of contents come out empty?
Because most of these tools insert a Word TOC field without filling it in, and Word does not update fields on its own. Open the file, right-click the contents and choose Update Field, or press F9. Kimi's file asks Word to do this when it opens; Gemini and Perplexity typed the list out instead, so it shows but will not update.
What about Grammarly?
It takes a .docx and gives one back, and it found every typo, the duplicated paragraph and the mixed numbering without being asked. But it is a checker, not a formatter - it will not make headings, a table or a table of contents, and its chat cannot see the document you have open; paste the text in and it rewrites the policy correctly, then promises a .docx it has no way to make. Use it on the words after one of the chat tools has done the structure.
Can ChatGPT or Claude edit a Word document?
Both can, on the free web apps - each takes a .docx, builds a new one with code, and hands it back. ChatGPT did it in two minutes with an empty TOC field. Claude spent twenty-three minutes and thirty-eight commands, most of them trying to pre-fill the table of contents with LibreOffice, ran out of tool calls for the turn, and delivered after one click on "Continue" - a good file, with a typed contents that has the right page numbers.