Priya read every email her AI worker wrote in its first week. By Friday she had approved 54 of them and changed four. That is a good way to start and a bad way to stay. Her rule, which this page spells out, is to review the first 50 emails, review anything going to a high-value account, and then let the worker act on its own.
So, should your AI worker ask first? Not by default. A worker that waits for your yes on every email is a drafting tool with a queue, and the queue is you. Turn review on at the start, for specific accounts, and in fields with rules about what you may say. Turn it off when the numbers below say it's earned that.
The two modes in the words the Settings page uses
In Rocketship the choice is one setting. Send on its own is the default: the worker finds the buyers, writes the email, sends it from your Gmail, reads and answers the replies, and books the meeting, and you see everything afterward. Show me each email first holds every email until you approve it. Nothing goes out until you say yes. The setting's own description warns that this is safer but your team waits on you.
There is a narrower version too. When you write a campaign yourself, a switch called "Let me press send on each one" holds only that campaign's emails in the Needs you list. The rest of the workspace keeps running. That switch is how you give one set of accounts more care than the others, and it's the tool for most of the cases below.
What review costs, besides your time
Two costs, and the first one surprises people. Delay: with review on, a reply from a warm lead at 4 p.m. sits until you open Needs you. If you check twice a day, the average wait is a few hours, and some replies wait overnight. A person who just wrote "yes, can you call me?" is the last person you want to keep waiting.
The second is rubber-stamping. After the 30th approval in a row, you're no longer reading. You're tapping. Review earns its keep only while you actually read, which is why a short, early window works better than permanent review. A queue you click through by habit protects you from nothing and still costs the delay.
Most advice online on this question is written for companies with dozens of reps and a brand team, and it says always require approval. Your situation is different. You are the brand team, you are also the person who has to do the approving, and at small-business volume one wrong email is a visible mistake you can fix, not a corporate incident.
The rule: review the first 50, then decide
Here is the rule in four lines.
- Turn review on for the first 50 emails of any new worker or new campaign.
- In those 50, count two things: how many you changed, and how many had a fact wrong (a service you don't offer, a price, a wrong industry term).
- If 50 in a row have at most two edits and none had a wrong fact, switch review off for that campaign.
- If you hit a wrong fact, fix the instructions first (see how to write instructions for an AI sales worker), then count 50 again.
The thresholds are mine, not a standard. Two edits in fifty means a typo or tone tweak now and then. A wrong fact is different in kind, because it means the worker believes something false about your business and will say it again.
Review time is small at this volume. Priya looked at about 30 seconds per email: 54 the first week, then 50 more after one correction, so 104 emails and about 52 minutes in total. She spent an hour to avoid reading emails for the rest of the year.
When to review first, and when not to
| Situation | Setting | Why |
|---|---|---|
| First 50 emails of anything new | Review on | You are checking whether the brief works |
| Accounts worth many times your average deal | Review on, for that campaign only | One bad email costs more than the review |
| Regulated advice or claims | Review on, and check your rules | The wording may be your legal responsibility |
| Routine outreach after a clean 50 | On its own | Speed matters and the risk is small |
| You changed the offer or the price | Review on for 20 emails | The old wording may linger |
High-value accounts get their own campaign
Priya sells commercial insurance. Most of her prospects are small shops with a $2,000 premium. A few are firms with fleets and a $30,000 premium. For the second group, one clumsy line is expensive, and the person is worth a note she wrote herself. So she keeps a short campaign for the big names with "Let me press send on each one" on, and the worker drafts, and she approves. Everything else runs on its own.
How do you define high-value? Pick a number. A reasonable line is any account worth ten times your average deal. Below it, the worker handles it. Above it, you read it. Having a number stops the ad-hoc "this one feels important" decisions, and it keeps the review queue short enough to read.
Regulated fields: you carry the wording
If you give financial advice, sell insurance, practice law or make health claims, what you say in a message to a stranger may be covered by rules you answer to, including advertising and solicitation rules from FINRA and the SEC, your state insurance department, or your state bar. This is general information, not legal advice, so check your own regulator and, if you have one, your compliance contact. In these fields keep review on for anything that describes returns, coverage, outcomes or qualifications. Plenty of outreach in these fields can still run on its own once it's confined to a plain introduction with no claims, but that's a call for your compliance contact, not for a blog post.
Switching review off without losing sleep
Turning it off isn't turning your eyes off. It moves the check from before sending to after. Open the Sent emails page once a day, filter to Replied and Bounced, and skim the last day's sends. The five-minute daily review walks through it. Keep a short list of triggers that put review back on: a complaint, a reply that says "I never asked for this," a wrong fact in the log, or a change to your price list.
Set the guardrails before the first send, not after. The day-one limits checklist covers send caps, hours, who it must never contact and what it must not promise. Those limits do most of the protecting. Review is the backup.
Priya's first two weeks
Days 1 to 3: review on, 54 emails read, four changed. Two changes were wrong terms (the worker wrote "workers' comp audit" where she offers general liability reviews), one was too long, one promised a "free quote" she'd rather call a "review." She added three lines to the instructions. Days 4 to 6: review on, 50 more, one edit for tone. Day 7: review switched off for the main campaign. The fleet-account campaign stays on review. Days 8 to 14: a five-minute log check each morning, and no changes needed.
If you want the reasoning behind a worker that acts, see hire an AI, not a tool and why replies need an answer in the first hour. The difference between an AI SDR that drafts and one that acts is in what an AI SDR really is. Because the sends, replies and bookings all live in one place (one login, one loop), the review queue is one list rather than three inboxes, and the AI workers page lists what each worker does.
Review is a setting you should use early and be willing to retire. Rocketship finds the buyers, emails them from your own Gmail, follows up, and books the meeting on its own, with "show me each email first" there when you want it. 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.