How to Write Cold Emails With AI: Prompts and Examples

AI can help you draft concise, relevant cold emails faster, but it cannot replace accurate research or your judgment. Start with a clear audience and a legitimate reason to contact them; then use AI to shape your notes into a message you verify before sending.

What You’ll Need

• A specific type of recipient and a clear purpose for contacting them.
• Verified details about the recipient or their organization, from appropriate public sources.
• An AI writing tool, plus a way to review, edit, and send messages.
• A truthful description of your offer and one low-pressure next step.

Step-by-Step

1. Define the audience and goal. Narrow the audience beyond a broad job title. For example, you might contact operations leads at small online retailers about reducing manual order-status work. Decide what one action you want: a brief conversation, permission to share a resource, or a reply about whether a problem is relevant. Avoid asking for several commitments at once.
2. Gather and check relevant context. Find a real, recent detail that explains why this person might reasonably hear from you, such as a stated business priority or a public announcement. Confirm the detail on a reliable source and record where it came from. Do not infer sensitive personal information, pretend to know the recipient, or use a detail simply because it is available. If you cannot establish a credible reason for the message, do not manufacture one.
3. Write a compact brief before prompting. Give the AI the audience, verified context, your offer, the problem you can address, the evidence you can honestly support, the desired next step, and a length limit. Mark facts it must not change. Leave out confidential customer information, private records, passwords, and other sensitive data. Check the tool’s data-handling terms before entering any non-public information.
4. Ask for a draft, not a finished campaign. Use a prompt such as:
   

   Write one concise cold email to [role] at [organization]. Goal: [one action]. Verified context: [fact and source]. Our offer: [plain description]. Relevant evidence: [specific, supportable proof, or “none”]. Keep it under 120 words, use a clear subject line, sound direct and courteous, and include one low-pressure question. Do not invent facts, results, familiarity, urgency, or personalization. If the context does not support a relevant message, say so instead of drafting one.

   Treat the word count as a guide, not a substitute for checking the actual message.

5. Check every claim and implication. Compare the draft with your brief and sources. Verify names, roles, dates, company details, figures, product capabilities, and any claimed results. Remove anything AI added that you cannot substantiate. Make sure the opening does not suggest a referral, prior conversation, or personal knowledge that did not happen. If a claim needs qualification, make that qualification clear.
6. Edit for a human voice and a single point. Replace generic praise and inflated business language with specific, plain wording. Explain the relevance to the recipient, not just what your product does. Cut repeated benefits, unnecessary background, and long introductions. Read the email aloud: it should sound like a respectful note a person might actually send.
7. Write a useful subject line and call to action. Ask AI for a few accurate, non-misleading subject lines, then choose one that reflects the email. Avoid fake “Re:” or “Fwd:” labels, false urgency, and vague clickbait. End with one clear, easy-to-answer request. For example: “Is reducing manual status updates a priority this quarter?” is more specific than “Can we connect?”
8. Review the recipient, timing, and sending requirements. Check the address and organization before sending. Follow the applicable rules for commercial outreach, privacy, and opt-outs in your location and the recipient’s location; requirements vary, so consult an appropriate source when uncertain. Use accurate sender details and honor requests not to be contacted. Do not blast an AI-generated list or keep messaging someone who has declined.
9. Send a small, reviewed batch and learn carefully. If you are testing variations, change one thing at a time, such as the opening or call to action. Review replies and complaints, not just opens, and stop or revise a message that produces confusion or negative responses. Keep personalization truthful and human-reviewed; automation does not make a poor-fit message appropriate.

Common Mistakes

• Invented personalization: AI may turn a thin clue into a confident claim. Use only context you verified, and remove it if it does not support the reason for writing.
• Overpromising: Do not let the tool create statistics, customer names, guarantees, or results. Include evidence only when you can substantiate it and have permission to use it.
• Generic copy: “I hope you’re well” followed by a broad pitch gives little reason to respond. State the relevant situation and why your offer may help.
• Too much selling: Long feature lists and multiple asks make the message harder to understand. Keep one main point and one next step.
• Sending without review: AI can produce incorrect names, awkward phrasing, or misleading subject lines. A human should inspect each message, especially personalized details.
• Ignoring contact preferences: Repeated follow-ups after a refusal damage trust. Respect opt-outs and check the rules that apply to your outreach.

Tips

• Ask the AI to identify unsupported claims and generic sentences after it drafts; verify its critique too.
• Give clear constraints: audience, purpose, tone, maximum length, and prohibited assumptions.
• Use a real detail to explain relevance, not as decoration. One accurate sentence is usually enough.
• Keep a record of the source for personalized facts so you can check them again before sending.
• Write the email for the recipient’s likely priorities. A polished pitch is not useful if the problem is irrelevant to them.
• Use AI for alternatives and editing, but make the final decision about accuracy, fit, and whether to send.

Troubleshooting

The draft sounds robotic: Provide a short example of your preferred tone without asking the tool to imitate a specific person. Request simpler wording, fewer adjectives, and shorter sentences, then edit it yourself.

The email feels generic: Revisit the audience and context. Add one verified detail and explain its connection to your offer. If you cannot make that connection honestly, choose a better-fit recipient or do not send.

The AI invents facts: Tell it to use only supplied facts and mark uncertain details, but do not rely on that instruction alone. Remove unsupported statements and verify every remaining claim against your notes and sources.

People do not reply: Silence can have many causes and does not prove that the copy is the problem. Check recipient fit, clarity, deliverability, and whether the ask is easy to answer. Test one respectful change at a time, and stop if responses indicate the outreach is unwanted.

The message is too long: Ask for a shorter version, then preserve the reason for contacting the person, the relevant offer, and the single ask. Cut generic introductions and extra claims rather than removing necessary context.

Frequently Asked Questions

Can AI write a cold email from scratch?
It can produce a draft, but it needs a clear brief and human review. Verify facts, fit, tone, and the recipient before sending.

What should I include in an AI prompt?
Specify the recipient, verified context, purpose, truthful offer, supported evidence, tone, length, and one call to action. Tell it not to invent details.

How long should a cold email be?
There is no universal ideal. Keep it as short as possible while making the relevance, offer, and next step understandable; remove material that does not help the recipient decide.

Is it okay to use AI to personalize emails?
It can help phrase personalization based on relevant, verified information. Do not imply a relationship that does not exist or use sensitive details inappropriately.

Should I automate sending AI-written emails?
Automation can scale mistakes as easily as good drafts. Review messages, check recipient fit and applicable requirements, and honor opt-outs before using any sending workflow.

What if I have no proof for a benefit?
Do not present it as a proven result. Describe the offer accurately, omit the claim, or gather reliable evidence before using it.

Related

• AI Job Displacement: What It Means and Which Tasks Are Exposed
• Agentic AI in FPGA Design Workflows
• Best Power Banks Over 20,000mAh: Top Picks and Trade-Offs