Most owners who let an AI worker send on its own make the same mistake in week two: they stop looking. The worker is working, replies arrive, nothing is on fire, so the dashboard stays closed. Then one wrong line goes to forty people before anyone notices. A five-minute review each morning catches nearly all of it, because the problems that matter show up in the first few emails, not the thousandth.

This is the routine, in order, with the time each step gets. It assumes the worker acts without asking first, which is the default in Rocketship. If you've turned on "Show me each email first," you're reading before sending and this page is your after-the-fact backup.

Why five minutes a day is enough

Reviewing everything the worker sends would defeat the point. You're not proofreading, you're sampling. A sample of three emails out of twenty tells you whether the instructions are working, and a quick look at replies and bounces tells you whether the list is. Anything that needs more time is a signal to change the instructions, not to read more emails.

Gareth runs a six-person commercial painting company and has a worker emailing facility managers. His mornings look like this, and it takes him four minutes.

Step 1: Is it running? (30 seconds)

Open Today. The line near the top says how the team is set to work (for most owners, "Your team sends on its own"). Under it is a card with a sentence like "14 of 20 sent today," and each inbox has a status chip: Ready, Paused, Limit reached for today, Reconnect needed, or a note that a new inbox is starting slowly to protect your email. The chip to act on is "Reconnect needed." It means Gmail dropped the connection (usually after a password change), and nothing will go out until you reconnect it. If the card says nothing was sent and you didn't pause it, find out why now, not on Friday.

Step 2: Read three sent emails (90 seconds)

Open Sent emails and look at yesterday. Don't scan the list, open three at random and read each one the way the recipient would. You're checking five things:

  1. Is the first line true? A line like "I saw your new location on Fifth Street" is only good if there is one.
  2. Is the name and company right, with no blank spots or mixed-up fields?
  3. Does it say anything you don't offer or can't do?
  4. Is it short enough to read in 15 seconds, with one question at the end?
  5. Are the opt-out line and your postal address in the footer?

One bad email out of three is a finding, not a fluke. Write it down, then see the fixes below.

Step 3: Read the replies that need you (60 seconds)

Filter Sent emails to Replied. Each reply carries a label: Interested, Asked a question, Proposed a time, Not interested, Out of office, Unsubscribed, Bounced, or Couldn't read the reply. Skip the ones that need nothing. Read two kinds in full: Asked a question and Couldn't read the reply. For each, check the worker's answer against what you'd have said. If a reply sounds off, the fix is almost always one line in the instructions, and the same line will prevent the next ten.

For why replies deserve the fastest attention in the system, see outbound that reads the reply.

Step 4: Count the bounces and opt-outs (45 seconds)

Filter to Bounced. A couple a day is normal. A rule of thumb many senders use is to keep bounces under about 2 in 100, so on a day with 20 emails, 3 or more bounces says the list has a quality problem. Look at what the bounced addresses have in common: an old source, one industry, one data provider. Then check Unsubscribed: those people should be out of every sequence the same day. Opens are a number you can ignore here. Why is explained in stop measuring opens.

Step 5: Meetings against sends (once a week, plus 5 minutes)

On Mondays, add the one comparison that tells you whether any of this works: meetings booked against emails sent. Gareth's last four weeks: 80 sent and 1 meeting, 100 and 2, 100 and 1, 100 and 3. That's 7 meetings from 380 emails, a bit under 2 percent. He doesn't know yet whether under 2 percent is good for facility managers, but the trend is up, and week four came right after he changed the opening line. When a meeting is booked, open that person and the timeline shows the email that started it. Over a month that tells you which opening works.

Three fixes for the problems you'll actually find

You'll see the same three problems over and over. Each has a one-line fix in the Rules and instructions box in Settings, which the worker treats as what it must always or never do.

The first line is generic or wrong

Add: "Open with something specific about their business only if you can state it as a fact from their website. If you can't, skip the personal line and start with what we do." The worker then falls back to a plain, honest opening instead of inventing one.

The email is too long or pushy

Add: "Keep every first email under 90 words. One question only. Never ask for a call more than once in a sequence." Look at the guide to writing instructions for how to pair this with examples in your own voice.

The wrong people are replying

If several replies say "I'm not the right person" or "we don't do that," the audience is off. Tighten the description of who you sell to and remove the title that keeps answering. Add titles you don't want to the rules: "Don't write to receptionists or assistants."

When to pause

Press Pause everything and read before resuming if: two emails in a day carry a wrong fact; a recipient says they never asked to be emailed and the list source is the reason; bounces spike well past a handful; or the inbox chip shows Reconnect needed and you can't fix it right away. Pausing costs a day of sending. A wrong fact repeated 80 times costs you the list. After you fix the instruction, test it the way the day-one limits checklist describes, then resume.

Reviews like this also tell you when review is no longer needed at all. The rule for switching it off is in autonomy vs approval, and the AI SDR explainer has the context on what a worker does. Everything lives in one workspace (one login, one loop), so the whole check happens on three screens: Today, Sent emails, and Needs you. The AI workers page shows what each worker covers.

A five-minute habit is what makes leaving the worker alone safe. Rocketship finds the buyers, emails them from your own Gmail, follows up, and books the meeting on its own. Plans that get you customers start at $24.99 a month (Launch), and both Launch and Frontier start with a 7-day free trial, card required.