How do I write a cold email with AI that doesn't read as AI?
Use AI for the research and the first pass, then cut the draft to 50 to 125 words: one line on who you are, a specific reason you picked this person, one small ask and an easy way to say no, with no stock opener or sign-off. You can do that with any chatbot and a good prompt, with saved instructions, or with a tool such as Ownhand that rewrites the draft in your own voice for a cold email.
A cold email is all the reader knows about you. Readers think less of you when you lean hard on AI. In a 2025 survey of 1,100 professionals, only 40 to 52 percent rated a manager as sincere when a message leaned heavily on AI. 83 percent did when it had light help.
AI is still useful here. It's good at finding facts about the person and turning your notes into a first pass. It's bad at what makes a cold email work: a real reason you picked this person, your own way of saying it, and stopping early.
What a cold email needs
Four parts, in this order, and nothing else:
- Say who you are in one line. Your name, your role and your company are enough.
- Say why this person. Give one specific fact you found, like a talk they gave, a job post, or a product they shipped. If the line would fit anyone else on your list, it's too vague.
- Make one small ask. A 15-minute call, a yes or no, the name of the right person. One ask, so the reader has one thing to answer.
- Give people an easy way to say no. A short line like "if the timing is wrong, no reply needed" makes it cheaper to answer.
The numbers back the short version. Boomerang looked at more than 40 million emails its users sent. Every length from 50 to 125 words got replies more than half the time. Emails with one to three questions were 50 percent more likely to get a reply than emails with none. Plain writing did best: emails at a third grade reading level got 36 percent more replies than emails written at a college level. Its data covers all kinds of email, so use it as a guide.
What gives an AI cold email away
Model drafts open and close the same way. Anyone who gets lots of cold emails spots them in the first line. These are the phrases Ownhand's email_cold occasion refuses:
| Phrase | What to write |
|---|---|
I hope this email finds you well | Nothing. Start with who you are. |
I wanted to reach out | Say why you are writing to this person. |
Quick question | Ask the question. |
pick your brain | Name the one thing you want to learn, and how long it takes. |
Would love to connect | Ask for a specific call on a specific day. |
Please don't hesitate to reach out | Drop it. A stranger who wants to reply knows how. |
Inside the email, watch for a vague benefit with no number, and words such as synergies
, streamline
or leverage
. For the full list of words and sentence shapes, see the phrases that make a message read as AI-written. Two real drafts, and what two writers made of them:
These examples are real output from the live API. Each AI draft was rewritten by a demo Hand: a fictional writer (an engineer, a founder, an ops lead, a designer or a consultant) built from a handful of samples.
AI draft: My name is Alex and I lead partnerships at Fieldnote. I'd love to explore potential synergies between our teams. Would you be open to a brief 15-minute call next week?
Demo Hand, consultant: Alex here, I lead partnerships at Fieldnote. Would you have 15 minutes for a call next week?
AI draft: I hope this message finds you well. I wanted to reach out because I believe our platform could help your team streamline invoicing and save up to 10 hours a week.
Demo Hand, consultant: Hi, our platform could save your team up to 10 hours a week on invoicing.
Both rewrites keep the name, the company, the number and the ask. The partnership one drops a vague line about the two teams. The invoicing one drops two lines of throat-clearing. Neither adds a fact the draft didn't have.
Step by step with any chatbot
- Collect the facts first: who you are, the one reason you picked this person, what you want, and one proof point like a customer or a number. Ask the AI to research the person, then check each fact yourself.
- Paste the facts with the prompt below. Ask for a short email up front. It works better than cutting a long one later.
- If the first line talks about the email instead of the reader, delete it. If the last line offers more help, delete it.
- Count the words. If it's over 125, cut the proof point down to one clause. If it's under 50, check that the reason you picked them is still there.
- Write the "why you" line yourself, or at least check it. A model can only guess why you care about this person. That's the line it most often gets wrong.
- Read it next to a cold email you wrote yourself and got a reply to. If yours is shorter and plainer, match that.
- Add a subject line of three or four plain words.
A prompt you can save and reuse:
Write a cold email from me using only the facts below. 50 to 125 words.
Order: who I am in one line, the specific reason I picked this person, one small ask, an easy way to say no.
No greeting line about hoping they are well. No line about reaching out. No closing offer to answer questions.
Plain words, short sentences, contractions, no bullets or bold. Do not add facts.
Facts:
If you write cold emails every week, save this in your custom instructions or in a project so you don't paste it each time. Guides for both are in making ChatGPT write in your voice and making Claude write like you.
Your options, compared
| Option | Cost | Sounds like you? | Enough when |
|---|---|---|---|
| Write it yourself, use AI only for research | Free | Yes | You send a few cold emails a week |
| Any chatbot with the prompt above | Free or your plan | Plain, but generic | You edit every email before it goes |
| Saved instructions or a project with your own past emails | Free or your plan | Closer, if the samples are good | One writer, one kind of email |
Ownhand with the email_cold occasion | $1 free, then about half a cent a rewrite | Built from your samples, and learns from what you send | Your agent drafts many kinds of message for you |
Follow-ups
Follow-ups are where AI drafts get lazy. A follow-up that only says just following up
or circling back
gives the reader nothing new. Name the earlier email in a few words, add one new thing, such as a result or a date, and repeat the ask. When you met the person first, start with where:
AI draft: It was truly a pleasure meeting you at the conference on Tuesday. I wanted to follow up on our conversation about the pilot and see if Thursday works for a call.
Demo Hand, consultant: Following up on our conversation about the pilot at Tuesday's conference, does Thursday work for a call?
The rules that apply to any cold email
Sounding human does not change the law or the spam filters. In the US, the FTC says the CAN-SPAM Act makes no exception for business-to-business email. A commercial email needs accurate sender details, a subject that matches the content, your valid postal address, and a clear way to opt out. You have to honor an opt-out within 10 business days. Each email that breaks the rules can cost up to $53,088. Other countries have their own laws, so check yours.
Gmail asks every sender to set up SPF or DKIM and keep the spam rate in Postmaster Tools under 0.3 percent. If you send more than 5,000 messages a day, you also need DMARC and one-click unsubscribe.
Where Ownhand fits
Ownhand is an MCP server. Your agent drafts the email, then calls the write tool with the email_cold occasion. The rewrite comes back in your Hand. It's built from 5 to 10 things you wrote. The occasion caps the email at 125 words, needs a greeting and a sign-off, and rules out Markdown and emoji. It puts who you are first, only if the draft says it, then why this person, then one small ask. It allows one hedge, on the ask, so the reader can say no.
Every rewrite is checked for the phrases in the table above, the listed words, and the stock sentence shapes. If a draft fails, the model gets the list of problems and tries again, up to two more times. Your agent shows you the email first. It sends nothing without your OK. When you send it, your agent reports the exact text, so the next cold email starts closer to how you write (see how Ownhand learns from what you send).
You don't need it if you write your own cold emails and only want a check. The steps and the prompt above work with any chatbot. It's worth the cost when an agent drafts many kinds of message for you each day and you want all of them to sound like one person.
Follow-up questions
How long should a cold email be?
Aim for between 50 and 125 words. Boomerang studied more than 40 million emails, and every length in that range got a reply rate above 50 percent. The study covered all kinds of email, so treat it as a guide for cold email and lean to the short end.
Is it fine to tell the reader I used AI?
You don't owe a disclaimer for help with wording. The facts and the reason you picked the reader have to be yours. One 2025 study found readers rated managers as much less sincere when messages leaned heavily on AI, mostly for messages meant to build a relationship.
Should I let an AI tool send cold emails for me automatically?
You have to read each one. Every commercial email has to meet the CAN-SPAM rules, and Gmail limits spam complaints for every sender. One wrong fact or a reply to the wrong person costs more than the time you saved.
Will Ownhand make up a reason I picked this person?
No. It keeps the facts in your draft and adds none. The cold email occasion puts who you are first only if your draft says it. The specific reason has to come from you or your agent's research.
Sources
Checked on 2 October 2026. If something here is out of date, email hi@thalientlabs.ai and we will fix it.
- Boomerang: 7 tips for getting more responses to your emails: 40 million emails; 50 to 125 words; 1 to 3 questions; 3 or 4 word subject lines; third grade reading level
- FTC: CAN-SPAM Act compliance guide for business: no exception for business email, postal address, opt-out within 10 business days, penalties up to $53,088 per email
- Google: Email sender guidelines: SPF or DKIM for all senders, spam rate under 0.3%, DMARC and one-click unsubscribe above 5,000 messages a day
- ScienceDaily: coverage of Cardon and Coman (International Journal of Business Communication, 2025): 1,100 professionals; 40 to 52 percent saw heavy-AI messages as sincere, against 83 percent for light help
- Wikipedia: Signs of AI writing: stock phrases and closers that readers associate with model text
- Ownhand source code: the cold email occasion: 125-word cap, required greeting and sign-off, no Markdown, the refused stock phrases