Agent-Run AgencyBuild log · Alex Rowan · Guides

Guide · Alex Rowan · 11 Oct 2026

AI cold email for local businesses: verify before you send

Build log

AI agents are good at the research half of cold email. They'll audit a hundred local business websites and draft a personal note for each one before you've finished your coffee. The risk is that every note makes a claim about someone else's business, and some of those claims will be wrong. This guide is how I handle that, with the real numbers from my first batch. No replies or clients are recorded yet, so it's about sending accurate email, not about results.

The near miss that shaped the process

My agents audited 154 small-business websites across 21 cities and turned the ones with real problems into 78 one-page mini-audits. Before the first send, a separate agent re-checked 30 of them against the live sites.

10 of the 30 cited a problem the business had already fixed. Seven times it was "no reviews or testimonials on the homepage" when reviews had since gone up. Twice it was a heavy homepage that had since been slimmed down. Once it was schema markup that was now in place. Two of the 30 had no confirmed problem at all.

A third of my first emails would have told an owner something false about their own website. Everything below exists because of that check.

Rule 1: email only published addresses

Of the 154 businesses, 92 publish an email address on their own website. Those are the only ones my agents will draft for. No guessed addresses, no scraped lists, no addresses pulled from somewhere the business didn't put them. Record where each address came from in the prospect sheet, so the answer to "how did you get my email?" is a URL.

Rule 2: make the claim specific and checkable

Vague claims ("your SEO could be better") can't be verified and don't get read. Each mini-audit names two specific issues from the business's live homepage, such as several H1 tags or no LocalBusiness schema, plus one fix they could hand to whoever runs the site. Specific claims are easy for an owner to check, which is the point, and also easy for an agent to re-check.

Rule 3: verify as late as possible, in a fresh session

The audit and the email are made at different times, and websites change in between. So verification is its own step, run right before each batch:

  1. Fetch every prospect's live homepage again.
  2. For each cited issue: confirmed, keep it. Not confirmed, remove it and regenerate the PDF.
  3. No confirmed issues left: skip the prospect entirely.
  4. Check the suppression list and drop any address or domain on it.
  5. Check each draft for sender identity, postal address, opt-out, an honest subject and at most one link.
  6. Write a short report: checked, passed, regenerated, skipped.
  7. Mark the batch "verified, awaiting owner approval".

Run it with a separate agent, not the one that wrote the audits. An agent checking its own work tends to agree with itself. A fresh session with a narrow brief ("confirm or drop each claim") catches more.

Rule 4: a person approves and sends

My agents never send email. They draft, the verifier checks, and I read each email before it goes. On day one the scripted send timed out partway through. Because sending was gated, nothing had gone out half-finished, and I sent the first 10 by hand. Ten at a time is slow, and that's fine for a first batch: you learn more from reading ten emails closely than from sending two hundred you've skimmed.

Rule 5: get the compliance basics right in the template

Local businesses in the US and Australia sit under different rules, CAN-SPAM in the US and the Spam Act in Australia, so the template carries the strictest common set:

  • A real sender name and business name
  • A full postal address, including any box number
  • A clear way to opt out, honoured quickly and recorded in a suppression list
  • A subject line that describes the email
  • No claims you can't back up

This isn't legal advice. Read the rules for the countries you send to, and treat the opt-out list as permanent.

Rule 6: keep the offer plain

The mini-audit is free and useful on its own. The owner can hand it to their web person and never reply. The email ends with a short, plain offer and prices stated up front, with no call required. Agents like to write urgency. Cut it. A local owner who gets a useful page and a clear price is more likely to remember you than one who got a countdown.

What the first week looks like

  • Businesses audited: 154 (97 US, 57 AU)
  • With a published email: 92
  • Mini-audits made: 78
  • Re-checked before sending: 30, with 10 corrected and 2 skipped
  • Emails sent: 10, by hand
  • Drafted for the next batch: 18
  • Replies: none recorded yet

How to brief the verifier

The verifier's brief is short on purpose. It gets the prospect list, the cited issues and one job: open each live homepage and, for every issue, answer "confirmed" or "not confirmed" with the evidence. It doesn't rewrite emails, suggest new issues or soften the wording. A narrow brief keeps it from drifting into being a second copywriter, and it makes the report easy to read in two minutes before you approve a batch.

Ask it to quote what it found. "Homepage now shows a reviews section with five entries" is far more useful than "issue resolved". When something is borderline, the rule is simple: if it can't show the evidence, the claim comes out.

Common mistakes with AI-drafted cold email

  • Letting the drafting agent check its own claims. It wrote them, so it will tend to confirm them.
  • Verifying at audit time instead of send time. Days can pass between the two, and that's long enough for a site to change.
  • Flattery and fake familiarity. "I love what you've built" from a stranger reads as automated, because it is. State the issue and the fix.
  • Too many links. One link at most, and plain text otherwise. It reads like a person wrote it, and it's kinder to the owner's spam filter.
  • Sending big batches too early. Start small, read the replies and the bounces, and fix the template before you scale anything.

Tracking what happens next

Keep a send log with the date, the prospect, the template version and the verifier report each email passed. When replies come in, you'll know exactly what each person was told. When someone opts out, they go on the suppression list the same day, and the next verification run checks it. My replies, bounces and opt-outs are still blank in the log because I haven't checked the sending mailbox yet. They'll go in as soon as I have them, zeros included.

A short checklist to copy

  1. Public addresses only, with the source recorded.
  2. Two specific, checkable issues per prospect.
  3. Re-check with a fresh agent right before the batch.
  4. Drop unconfirmed claims; skip prospects with none left.
  5. Suppression list checked every time.
  6. A person reads and sends.

If you only take one thing from this guide, make it the late re-check. It costs a few minutes per batch, it needs no special tools, and it's the difference between an email that helps an owner and one that tells them something untrue about their own business.

The agents make the research fast. The verifier and the approval step make it safe to put your name on.

Free

Get the AI Agent Team Playbook

The five agent briefs, the approval gates and the verifier routine from this build, in a free 9-page PDF. No promises of rankings or income.

Get the free playbook