Polluxa

AI Cold Email: How It Works and How to Keep It Personal

AI can research prospects and draft cold emails in seconds. Here is where that helps, where it backfires, which deliverability rules apply, and how to keep each email personal.

AI cold email means using AI to research prospects, write first-touch sales emails and follow-ups, and help handle replies. It helps most with research and first drafts, and hurts most when teams use it to send more generic email, faster. The teams that do well use AI to write one relevant email for one person, then check it before it goes out.

How do AI cold email tools work?

Most AI cold email tools follow the same five steps: build a list of prospects that match your ideal customer, research each one, draft an email, send it with follow-ups on a schedule, and sort the replies. The difference between tools is how good the research is and how much control you keep.

  1. Find prospects. You describe your ideal customer profile: industry, company size, region, job titles. The tool finds matching companies and people, or you import a list.
  2. Research. The AI reads what it can find about each person and company: their website, recent news, job posts, public posts.
  3. Draft. It writes an email using that research plus what you tell it about your product, your proof and your offer.
  4. Send and follow up. Emails go out from your mailbox on a schedule, with follow-ups when there is no reply.
  5. Handle replies. It sorts replies (interested, not now, wrong person, unsubscribe) and drafts answers for you to review.

Where does AI help in cold email?

AI helps with the slow, repetitive parts: researching each prospect, finding a real reason to write, drafting a first version, and keeping follow-ups on time. That frees reps to spend their time on replies and calls. It also helps small teams write consistently without a full-time copywriter.

  • Faster research. If your reps spend ten minutes per prospect reading websites and news, AI can do that first pass while they work on replies.
  • A real reason to write. Good AI tools pull a specific trigger, such as a new office, a hiring push or a product launch, and connect it to your offer.
  • First drafts. Editing a decent draft is faster than writing from a blank page.
  • Follow-up discipline. Most cold deals need more than one email. Software never forgets to send the second one.
  • Reply sorting. "Not interested," "talk to my colleague" and "unsubscribe" each need a different action. AI can route them so nothing sits in an inbox.

Where does AI hurt cold email?

AI hurts when it is used to send more of the same email instead of a better email. Generic copy, invented "personal" details, and higher volume all lead to the same place: fewer replies, more spam complaints and a damaged sending domain that also hurts your normal business email.

  • Generic writing. Stock openers like "I hope this finds you well" or "I came across your profile and was impressed" read as automated, and buyers learn to skip them.
  • Wrong facts. AI can misread a website or mix up two companies. One wrong "personal" detail does more harm than no detail at all.
  • Fake personalization. Mentioning someone's university or a post from two years ago is personal but not relevant. Relevance is what earns a reply.
  • Too much volume. When writing is free, teams send more. Google's own guidance is to start with a low volume and increase it slowly, and to avoid sending in bursts.
  • Spam complaints. Every complaint raises the spam rate that Gmail and Yahoo measure for your mail.

What deliverability rules apply to AI cold email?

Gmail and Yahoo publish clear rules. Every sender needs SPF or DKIM and a spam complaint rate below 0.3%. Senders who reach about 5,000 messages a day to personal Gmail accounts also need DMARC, an aligned From domain and one-click unsubscribe on marketing mail. AI-written email gets no exception.

RequirementGoogle (Gmail)Yahoo
SPF or DKIMAll sendersAll senders
Spam rateBelow 0.3%; aim for below 0.1%Below 0.3%
SPF and DKIM bothBulk sendersBulk senders
DMARCBulk senders; policy can be "none"Bulk senders; at least p=none, must pass
From domain aligned with SPF or DKIMBulk sendersBulk senders
One-click unsubscribeBulk senders, on marketing and subscribed mailBulk senders, on marketing and subscribed mail
Honor unsubscribesWithin 48 hoursWithin 2 days

A few details matter for sales teams:

  • Google's bulk threshold is close to 5,000 messages in 24 hours to personal Gmail accounts, counted across your whole primary domain. Once you hit it, Google treats you as a bulk sender permanently.
  • Google also asks for valid forward and reverse DNS and TLS. Most email providers handle these for you, but check.
  • Google started ramping up enforcement on non-compliant mail in November 2025.
  • These rules are written for personal Gmail and Yahoo mailboxes. Most B2B email goes to company domains, but these are the clearest public rules from big mailbox providers, and meeting them costs little.

Google's guidance also says not to buy email lists and not to send to people who did not sign up. Cold email, by definition, goes to people who did not sign up. That is the honest tension. The way through it is low volume, clear relevance to the person's job, an easy way to opt out, and stopping the moment someone asks.

Check the law in your buyers' countries too. In Canada, for example, CASL requires consent before sending commercial electronic messages. Consent can be implied when someone has conspicuously published their business email, has not said they refuse unsolicited messages, and your message relates to their role. Every message must identify you and include an unsubscribe option. Ask a lawyer if you are unsure.

How do you keep AI cold emails personal?

Give the AI one clear reason this person should care, check every fact it uses, and keep the email short enough to read on a phone. Then read each one before it goes out, at least until you trust the output. Personal means relevant to their job right now, not a compliment about their profile.

A simple checklist for every AI draft:

  1. One reason, stated early. Why this person, why now? If you cannot answer in one sentence, do not send.
  2. Every fact checked. Click the source. If the AI says they opened a Riyadh office, confirm it.
  3. One small ask. "Worth a 15-minute call next week?" beats a list of options.
  4. Your voice, not the model's. Cut filler openers, cut adjectives, and read it out loud.
  5. No tracking tricks or false "Re:" subjects. They break trust and can trigger complaints.
  6. An easy way out. A line like "If this isn't relevant, just say so and I won't follow up" is polite and lowers complaints.

What does this look like for a real team?

Take a fictional 3-person sales team in Toronto that sells maintenance software to food manufacturers. They define their ideal customer as plants with 100 to 1,000 staff in Ontario, and pick plant managers and operations heads as their contacts.

They use AI to research each plant and draft the first email around one trigger: a new production line, a hiring post for maintenance technicians, or a recent food safety audit mentioned in local news. A rep reads every draft before it goes out. They send from one domain with SPF, DKIM and DMARC set up, keep daily volume modest and steady, and stop all emails the moment anyone replies or asks to be removed. Because they sell into Canada, every email names the company and includes an unsubscribe line.

Where does Polluxa fit?

The Polluxa Email Agent sends from your own Gmail. It researches each prospect, writes one email for one person, follows up when they go quiet, and helps you answer replies. Every thread is saved in Polluxa CRM, so the next rep sees the whole history. If you also need the list itself, Polluxa can find companies and decision-makers that match your ideal customer profile, and lead enrichment fills in the details.

Many other tools write AI cold email. Judge any of them by the same test: does each email read like it was written for one person, and does the tool help you send less, better email?

FAQ

Is AI cold email effective?

It can be, when AI handles research and drafting and a person checks relevance and facts. It fails when it is used to send more generic email. The result depends on your list, your offer and your sending discipline more than on the writing tool.

Will AI-written emails go to spam?

Not because AI wrote them. Mailbox providers judge your authentication, your sending patterns and how recipients react. Set up SPF, DKIM and DMARC, keep volume steady and low, and keep your spam rate well under Google's 0.3% limit.

How many AI cold emails can I send a day?

Google and Yahoo do not publish a safe daily number for cold email. Google's guidance is to start low with engaged recipients and increase slowly, at a consistent rate. If your team sends 30 emails a day per mailbox and complaints stay near zero, grow slowly from there.

Do I need to disclose that AI wrote the email?

Google's and Yahoo's sender rules do not mention AI. What matters is that the email is accurate, identifies you honestly and gives an easy way to opt out. Check the rules in your buyers' countries, since some laws set content requirements.

Should a person review every AI cold email?

At the start, yes. Review every draft until you know where the AI makes mistakes for your market. After that, many teams review a sample each day and always review emails to senior buyers or key accounts.

Can AI handle replies to cold emails?

It can sort replies and draft answers, which saves time on "not now" and "wrong person" responses. Interested replies should go to a person quickly, because that is where deals start.

Want AI cold email that researches each prospect and writes from your own Gmail? Start the 14-day free trial.

More Blogs

No More Blogs