Last updated: 5 November 2026
The best way to write a personalised B2B cold email with AI is to work in steps: summarise the prospect’s business, choose one reason to get in touch, draft a short email, edit it so that it sounds like you, and check every fact before sending. Each step has its own prompt, and this guide gives you all of them.
The prompts work in ChatGPT, Claude, Google Gemini and Microsoft Copilot. Replace the words in square brackets with your own details. The example emails use an invented company so that you can see the shape of the result, and they do not describe a real customer. Before you send anything, read our PDPA and cold outreach checklist, because who you email matters as much as what you write. For the full sales process, see our practical guide to AI for sales.
Why do most AI cold emails read as generic?
Most AI cold emails read as generic because the prompt gives the assistant nothing specific to work with. If you ask for a cold email to a marketing director, you get an email that could go to any marketing director. A specific prospect, a specific reason and a clear next step give the draft something real to say.
So the prompt needs the prospect’s business, a reason you are writing now, what you offer and the one action you want the reader to take. The steps below collect each of these in turn.
Step 1: How do you research a prospect with AI?
Ask the assistant to summarise the prospect’s business from information you give it, and to suggest what problems the business may have that you can help with. Paste the text from their website or a recent announcement yourself, so that the summary rests on real material and not on guesses.
You are a B2B sales researcher. Below is text from a prospect company’s website. Summarise what the company does, who it sells to and how it makes money in four sentences. Then list three problems a company like this commonly has that [my product or service] could help with. Mark any problem you are guessing at as a guess.
[Paste website text here]
The last instruction matters. It makes the assistant separate what the text says from what it assumes. Check the summary against the website before you use it. Company information is enough for this step, so leave personal details about individuals out of the prompt.
Step 2: How do you choose the angle?
Choose one reason to write and one outcome for the prospect. Ask the assistant for three possible angles, then pick the one you can back up with something real, such as a result you can share or a change in the prospect’s business that you noticed.
Based on the summary above, suggest three angles for a first email from [my company] to the [job title] at this company. For each angle, give one sentence on why it might matter to them now and one thing I would need to be able to prove. Do not invent facts about the prospect.
Step 3: How do you draft the email?
Give the assistant the prospect, the angle, what you offer, the length and the call to action in one prompt. Ask for a short draft with a plain subject line, and tell it what to avoid. Short emails are easier to read on a phone, so set a word limit.
You are writing as [my name], [my role] at [my company] in Singapore. Write a first cold email to the [job title] at [company]. Angle: [chosen angle]. What we offer: [one sentence]. Proof I can share: [one real example, or leave blank]. Ask: a 15-minute call. Tone: polite, direct and plain. Length: under 120 words. Give me two subject lines. Do not use buzzwords, do not flatter the reader, do not make claims I have not supplied, and do not mention anything about the prospect that is not in the text I gave you.
Here is what a weak prompt and a strong prompt produce in an invented example. A Singapore firm that sells scheduling software to clinics is writing to the operations manager of a clinic group.
From a weak prompt (a cold email to a clinic manager):
Subject: Boost your clinic’s efficiency
Hi [Name], I hope this email finds you well. We help clinics streamline operations with our innovative scheduling platform. Would you be open to a quick chat to explore how we can help?
From the structured prompt above:
Subject: Booking clashes at [clinic group]
Hi Ms Lim, your website says patients can book online, by phone or at the counter. When bookings arrive through three channels, double bookings are hard to avoid. We make scheduling software that puts all three in one calendar. Would a 15-minute call next week be useful? If not, tell me and I will not follow up.
The second email works better because it uses something from the prospect’s own site, names one problem, makes one request and offers an easy way out.
Step 4: How do you make the email sound like you?
Ask the assistant to edit the draft against samples of your own writing. Give it two emails you have written yourself, then read the result aloud and change anything you would not say to the person’s face.
Here are two emails I wrote that sound like me: [paste]. Rewrite the draft below to match that voice. Keep it under 120 words. Remove phrases such as ‘I hope this email finds you well’, ‘game-changing’ and ‘touch base’. Keep every fact exactly as it is.
Cut the opening line if it carries no information. Most cold emails improve when the first sentence is the reason you are writing.
Step 5: What should you check before you send?
Check every fact about the prospect, confirm that you are allowed to contact them, and make sure the email gives a way to opt out. The AI draft is a starting point, and you are the one who sends it.
- Facts. Verify every claim about the prospect’s business against their own website or announcements.
- Claims about you. Remove any result, customer name or number that you cannot support.
- Permission. Check the list source, the Do Not Call rules and the Spam Control Act points in our PDPA and cold outreach checklist.
- Identification and opt-out. Include your name, company and contact details, and a simple way to say no.
- Tone. Ask yourself whether you would be comfortable if the prospect forwarded the email to a colleague.
What does a follow-up email prompt look like?
A follow-up should add something new, such as a short example or a useful resource, and keep the same request. Give the assistant your first email and ask for a brief follow-up that adds one new piece of value and does not repeat the first message.
Here is my first email. No reply came after [number] days. Write a follow-up under 70 words that adds one new piece of information the reader may find useful, keeps the same request and makes it easy to say no. Do not pressure the reader or suggest that they have ignored me.
Decide in advance how many follow-ups you will send and when you will stop. If someone asks you to stop, remove them from your list straight away.
What mistakes should you avoid with AI cold emails?
- Inventing facts. Assistants can state invented details about a company with confidence. Give them the source text and check the output.
- Sending the same email to everyone. Personalisation only works if each email reflects its own prospect. Draft by segment, then edit each email.
- Writing too much. A cold email asks for one small thing.
- Hiding the reason. Say early why you are writing and what you are selling.
- Skipping the edit. Readers notice text that nobody has read.
Learners in our two-day Mastering AI for Sales Growth and Team Dynamics course practise outreach like this with live prompting on their own sales scenarios, then build a working AI prospecting flow across email, LinkedIn and CRM touchpoints.
Frequently asked questions
Which AI tool is best for writing cold emails?
ChatGPT, Claude, Google Gemini and Microsoft Copilot can all write a good first draft. The quality depends more on your prompt and your edit than on the tool, so use the one your team already has. Our comparison of the four explains the differences.
How long should a B2B cold email be?
Short enough to read on a phone without much scrolling. The examples in this guide stay under about 120 words, with one reason for writing and one request.
Can AI personalise emails at scale?
It can speed up the research and the first draft for each prospect. The more you send, the more important it becomes to check facts and lists, because mistakes multiply with volume. A person should review every email before it goes out.
Do I need to worry about the PDPA when I use AI for cold email?
Yes. The PDPA, the Do Not Call rules and the Spam Control Act can all apply, depending on whom you contact and how. Our PDPA and cold outreach checklist explains the main points. It is practical guidance and is not legal advice.
What should I do if nobody replies?
Send a short follow-up that adds something new, then stop after the number of messages you decided in advance. If the contact asks to be removed, remove them.
Where does this guide fit in our AI for sales series?
This guide is part of our series on AI for sales. Start with the practical guide to AI for sales, then read our guide to building an AI sales coach for objection handling and the article on AI for sales prospecting.
Practise AI outreach with a trainer
QD Academy’s two-day Mastering AI for Sales Growth and Team Dynamics course is taught by practitioners and funded for eligible learners. Check the course page for the latest dates, fees and funding.