whichforwhat
Work and productivity

Best AI for writing emails

A payment chase that has to stay friendly, a no to a supplier that has to stay warm, an apology that must not grovel - and the one rule that matters, say what I told you and nothing else. We gave nine assistants the same three situations with fifteen facts planted in them and read every line that came back.

People also look for this as email writing, business emails, professional emails, email assistant or write my email.

Our picks, side by side

Rated for this task · 0–3
Measured
Quality
Effort
Cost
Free tier
From
Why
★ Best overall

ChatGPT

$8/mo · free tier: usable
Measured15 of 15 facts · 0 added · 3 of 3 under 150 words · ~30 s
Quality
Effort
Cost
Free tierUnlimited text chat on the default model, with limits on uploads, images, voice and deep research
From$8/mo
Why

Every fact, nothing else, in thirty seconds - and each email arrives as a card with the subject in its own field and a send button, so it is a draft, not a chat message you copy out. The tone is plain and friendly as asked, and the no to the supplier is the gentlest of the nine that still says it: "I understand, but I would need to look at other suppliers." The subject lines are terse; you may want to lengthen them.

Best if you want it to sound kind

Claude

$20/mo · free tier: usable
Measured15 of 15 facts · 1 added · 3 of 3 under 150 words · ~40 s
Quality
Effort
Cost
Free tierChat with web search and file uploads, with a usage cap per five-hour session
From$20/mo
Why

The warmest of the nine without going soft - it opens with a line for Miguel, signs the supplier email "Warmly", says sorry to Inês twice and still gets every fact in under 110 words. One thing it added on its own, a reason why the three-week lead time matters, is the reason Ana would probably give; read it before you send. The no is clear: "I'd have to start looking at other suppliers."

Best if you want it to sound like you

DeepSeek

Free · free tier: usable
Measured15 of 15 facts · 1 added · 3 of 3 under 150 words · ~30 s
Quality
Effort
Cost
Free tierThe full chat app with web search and file uploads; there is no paid consumer tier
FromFree
Why

The shortest emails that still read as a person - 65 to 87 words, every fact, and one line in each that carries what Ana meant without a flourish: "I'd rather not add it", "I'd much rather continue with you", "Again, sorry for the trouble." It did put one reason in her mouth, that the lead time "matters a lot for my shop", which is true of anyone asking for it. No cards, no offers to revise, just the three emails.

Evidence

15 of 15 facts; 0 added; 3 of 3 under 150 words (78, 86, 95); 3 subject lines; 3 signed as asked; ~30 s.

measured number · · Read the record · All 14 artifacts

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
Ratings support the verdict; they do not compute it. Prices checked on the date shown in the byline.
ToolQualityEffortCostFree tierMeasuredFrom
ChatGPTFree: Unlimited text chat on the default model, with limits on uploads, images, voice and deep research15 of 15 facts · 0 added · 3 of 3 under 150 words · ~30 s$8/mo
ClaudeFree: Chat with web search and file uploads, with a usage cap per five-hour session15 of 15 facts · 1 added · 3 of 3 under 150 words · ~40 s$20/mo
GeminiFree: The default model with usage limits, plus Gemini inside Maps, Gmail and Docs on a free Google account15 of 15 facts · 1 added, a motive · 3 of 3 under 150 words · ~90 s$19.99/mo
PerplexityFree: Unlimited basic searches with a weekly cap on Pro-model searches15 of 15 facts · 0 added · 'will' softened to 'may' · ~20 s$20/mo
Microsoft CopilotFree: The Copilot app on web, Windows and mobile with usage limits15 of 15 facts · 0 added, 1 stock line · 3 of 3 under 150 words · ~1 min$19.99/mo
DeepSeekFree: The full chat app with web search and file uploads; there is no paid consumer tier15 of 15 facts · 1 added · 3 of 3 under 150 words · ~30 sFree
GrokFree: Chat with web and X search within generous limits, no card required15 of 15 facts · 0 added · 2 of 3 signed, cut off · folded answer$30/mo
QwenFree: The full chat app; there is no paid consumer tier15 of 15 facts · 0 added · 53–58 words, no warmth · ~3 minFree
KimiFree: Unlimited basic chat, file upload and web access on the free plan15 of 15 facts · 2 added, rhetoric · 3 of 3 under 150 words · ~1 min$19/mo
GrammarlyFree: spelling, grammar and tone checks on everything you type, a document editor with .docx upload and export, 100 AI prompts a month15 of 15 facts · 2 added · 'will' softened to 'may' · 3 of 3 under 150 words · by hand$12/mo

How we tested

One message to ChatGPT, Claude, Gemini, Perplexity, Microsoft Copilot, DeepSeek, Grok, Qwen and Kimi on their free tiers, memory off, in a fresh chat. It described a small Lisbon shop and three emails to write - an overdue invoice with a late fee to waive, a supplier's 12% rise to counter at 6% with a condition, a customer's broken bowl with two remedies - each under 150 words, with a subject line, signed by the owner, and using only the facts given.

Every name, number and date is invented. Each email carries five facts that must appear, fifteen in all, and a script checked them along with the subject lines, the body word counts and the sign-off. Then we read each email for anything the brief did not say - a motive, a claim, a policy, a stock phrase - and compared how each tool worded the hardest sentence, that Ana will have to look at other suppliers.

Grammarly, the Roster's writing Specialist, is not a chat, so it got the brief through the "Ask AI" box of a new document in its editor - by hand, since that editor cannot be automated - and wrote the three emails straight into the document.

All ten got all fifteen facts and stayed under 150 words. Five added nothing. Gemini gave Ana a reason for leaving her supplier that she never stated; Kimi argued her case with two claims of its own; Claude and DeepSeek each added one reason why the lead time matters. Perplexity softened "will have to" into "may need to". Grok's answer was hidden in a folded panel and stopped before the last sign-off; Qwen's three emails ran 53 to 58 words with no warmth on the hardest one. Grammarly softened the no to "may" too, and gave the supplier a "While I appreciate your situation" the owner never said.

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 writing emails?

On this test, ChatGPT - every fact, nothing added, friendly, and each email delivered as a send-ready card in half a minute. Claude writes the kindest version and DeepSeek the plainest; both slipped in one reason the owner had not given. All ten, Grammarly included, got the facts right, so the choice is about voice and what you are willing to check.

Will they add things I did not say?

Some will. Gemini explained why the shop would leave its supplier, "to manage our costs and inventory timelines", which the owner never said. Kimi wrote that a 12% rise "is more than we can absorb". Copilot added the stock line "if the payment has already been made, please disregard". None added a discount, a deadline or a threat, and none changed a number. Read the sentences that argue, not the ones that state.

Which one is best for a firm but polite no?

Copilot said it most plainly - "I cannot accept that level of increase" - in the most formal voice of the nine. ChatGPT and DeepSeek said it gently and clearly. Perplexity said "I may need to look at other suppliers", which is weaker than the owner asked for. Kimi wrote a paragraph making the case, which reads well and goes beyond the brief.

Is Grammarly any good at writing emails, not just checking them?

Its Ask AI box wrote three usable emails with every fact and put them straight into the document, on the free plan's 100 prompts a month. It softened the hardest line to "I may need to consider other suppliers" and added a sympathy the owner had not offered, so it sits mid-table. Its real home is inside Gmail or Word, checking an email you wrote yourself, which this test did not measure.

Can I just send what it writes?

After one read, yes, for the seven that delivered whole. ChatGPT's cards have a send button. Check the subject lines - ChatGPT's and Qwen's are two or three words - and the sign-off, which Grok dropped from its last email.

Related tasks