Cold Email With AI: Research, Draft, and Review Outreach

Use AI for cold email by supplying a verified reason to contact someone, asking for a short draft, and checking every claim before a person approves sending. AI can help organize the message; it cannot establish that a stranger wants it or that your offer fits their business.

Keep the reusable instructions separate from recipient data. fenbs can hold approved communication prompts, audience assumptions, and lessons from reviewed outreach so the next assistant starts from the same rules. Your email system should remain responsible for sending and maintaining the list of people who must not receive more messages.

How do you start a cold email without inventing relevance?

1. Define one audience and one useful offer. Write down who might benefit, what problem you can actually address, and who falls outside the audience.
2. Find a legitimate, specific reason for contact. Open the company source yourself and record what it says, its date, and the source URL.
3. Separate observation from assumption. A company announcing a new location is observable; its team struggling with expansion is an unconfirmed guess.
4. Give the assistant only the approved facts, a restrained tone, and one possible next action. Ask for a draft and a list of unsupported assumptions.
5. Review the recipient, message, opt-out handling, and sending setup before approving the email in your email system.

Say a hypothetical design studio offers accessibility reviews for online stores. A published redesign announcement may justify a brief introduction. It does not justify claiming the company has accessibility failures, lost sales, or an unhappy design team. If there is no useful reason to contact this business, improve the audience choice before improving the wording.

What facts should you give the assistant?

Prepare a small research card rather than a biography. Include the business activity relevant to your offer, the exact public observation you checked, your verified capability, and an appropriate contact role. Avoid importing unrelated personal details just because you can find them.

Choose approved source material deliberately. Do not ask the assistant to turn a list of names into intimate-sounding observations. Keep the reusable audience criteria in your AI work routine, while recipient details stay in an approved system with appropriate access.

What prompt produces a useful first draft?

Request a modest draft with visible unknowns. A word limit can help remove sales padding, but factual limits matter more. The assistant should be allowed to tell you that the evidence is too weak to write a relevant message.

Draft one first-contact business email for human review.
Audience: [business role and reason the offer fits].
Verified observation: [fact, source URL, and date checked].
Our verified offer: [deliverable we can provide].
Unknowns: [do not fill these in].
Ask: [one small, optional next action].
Use a plain subject and about 100 words. Do not imply a prior relationship, invent praise, diagnose a problem, or promise results. Preserve required sender and opt-out text supplied below. After the draft, list every statement that still needs checking. Do not send.

For the studio example, the core could read: “Your announcement describes a new online store. We review selected store pages and return a practical accessibility issue list. Would an outline of that review be useful?” This is illustrative message text, not a complete send-ready commercial email. Add your approved sender information and required disclosures before use.

Save the version a person approves, together with its intended audience and exclusions, in fenbs Skills and prompts. An approved prompt is a starting instruction; each generated message still needs review.

What sending checks matter in the United States?

The FTC CAN-SPAM business guide↗ (opens in a new tab) says the law covers commercial messages, including business-to-business email. Its requirements include accurate sender information and subjects, identification as advertising, a valid physical postal address, an understandable opt-out method, and timely handling of opt-outs. Use that guide to check the full requirements for your message rather than treating an AI draft as a compliance review.

A technically deliverable message also needs an appropriate sending setup. Google’s sender guidelines↗ (opens in a new tab) describe authentication and other requirements for messages sent to personal Gmail accounts, with additional requirements for bulk senders. Have the person responsible for your email service check the requirements that apply; polished copy does not fix an unauthenticated domain.

• Check the actual To, From, Reply-To, and subject fields, not just the message body.
• Confirm the recipient is eligible for this outreach and is absent from your current suppression list, the record of people you must not email.
• Test that opt-out requests reach the responsible person or system and prevent further outreach.
• Review the destination of every link and remove tracking or attachments that your approved process does not need.

Other rules may apply to your organization, recipients, and locations. Have a qualified reviewer resolve uncertainty before sending. The separate guide to whether AI assistants can send email covers sending authority; drafting permission alone should not settle that decision.

How should you review a small outreach test?

Review a small, deliberately chosen set before considering broader use. Check whether each email has a truthful reason to exist, whether the ask is easy to decline, and whether the recipient can understand who is contacting them. A smooth sentence is not evidence of a good audience choice.

Record what you actually observed: replies, complaints, mistaken assumptions, or opt-out failures. Do not label silence as interest or treat an open notification as proof that someone read the offer. Change one meaningful element at a time so a later reviewer can understand what changed.

Keep a lesson such as “our location-based assumption did not establish a need” rather than copying private replies into a general knowledge record. This is the useful role of an AI brain: preserving the reasoning that improves the next task. For general composition outside outreach, use the AI email writing guide.

How to do this with fenbs

1. Create a free account and add an outreach project with its audience, offer, and human reviewer.
2. Store the reviewed drafting instructions in Skills and prompts. Record a person’s continuing communication requirements in Decisions and rules.
3. Add a task for the next message review. State what evidence is ready, what remains unverified, and which person approves sending.
4. Ask a connected assistant to read the project context before drafting, then save the review result and next step in Where we left off.

fenbs does not send emails or maintain your suppression list. If you only need to compose one message, an email editor may be enough; fenbs becomes useful when approved prompts and lessons need to survive the next assistant session. Explore more practical workflows in AI at work.