Cold email rules we enforce on every AI written message

Every email the Sellinger agent writes follows hard rules that outrank style: personalise only with facts in the lead data, never invent a price, number, date or availability, never promise a call or an offer by a deadline, answer price questions only with approved numbers or hand over to a person, and treat a no or a stop as the end of the thread. Around the text sit system checks: a suppression check before every send, daily limits per mailbox and machine mail detection.

Most email advice is about style: short subject lines, one ask. Style is easy to copy. The rules that protect you and the recipient are about facts, promises and stopping, and they are the ones an AI writer is most likely to break if nobody has written them down. This post lists the ones we enforce. We describe only rules that are live today.

Which rules apply to every email?

1. Facts only from the lead data

The agent personalises with what is present in the lead's data and nowhere else. It never describes itself with the recipient's title or company. A line such as "I saw you lead growth at X" is allowed only if the data says so. An invented personal detail is worse than no personalisation.

2. The greeting names the recipient

The greeting names the person you are writing to, never the sender. Writing the sender's own name where the recipient's belongs is the worst error an email can contain, and we rank it that way in the rules.

3. No invented prices, numbers, dates or availability

The agent never invents a price, a number, a date or an availability. It never promises a phone call, or an offer by a deadline. If it does not have the fact, it does not write one.

4. Price questions get approved numbers or a person

When the lead asks for a price, a quote or availability, the agent answers with the approved numbers from its knowledge. If they are not there, it does not guess and does not promise. It flags the lead as needing attention so that a person answers. This is the allowed use of that flag.

5. A no or a stop ends it

A refusal ends the exchange with at most one polite line and no further pitch. A reply such as STOP or unsubscribe, in English, Bulgarian or German among other forms, puts the contact on the team's suppression list, the agent stops and the lead is closed out. A weak or unclear phrase never writes the permanent record: the agent stops and a person decides.

6. A follow up keeps the same offer and price

Later emails in a thread keep the same offer and the same price as the earlier ones. A follow up is not a place to improvise a discount.

7. Plain formatting

No markdown in the email body. A subject line on every new email. No sign-off, name, title, phone, company line or booking link in the body, because the HTML signature is added automatically. The message ends on its last content sentence. And no dashes used as punctuation: a comma or a new sentence instead.

Which checks happen around the text?

The text rules sit inside a system.

• Suppression check. Before every send the platform checks the team's suppression list and refuses the send if the contact is on it, or if the check cannot be completed. It fails closed.
• Daily limits. Each mailbox has a limit the agent cannot exceed. The default is 50 sends a day, with an optional warm up that starts at 5 a day and doubles every 3 days up to the limit. See daily limits and pacing.
• Schedule. Sending follows the days, hours and time zone you set. With no schedule, a quiet window of 01:00 to 08:00 local time applies.
• Machine mail. Automatic replies from helpdesks, no reply senders, satisfaction surveys and empty messages are marked, never count as a human reply, are not answered, and the lead goes to a person with a named reason. See keeping AI agents out of machine mail.
• Unsubscribe. An unsubscribe line and the standard one click unsubscribe headers (List-Unsubscribe and List-Unsubscribe-Post) are available as a per mailbox setting. It is off by default, so check it for each mailbox. Google and Yahoo ask bulk senders for one click unsubscribe, as explained in Google and Yahoo bulk sender rules.
• Erasure. An erasure function takes an email address, phone number or LinkedIn address, anonymizes the matching leads in the workspace and adds the identifiers to the suppression list, so the person cannot be imported and contacted again.

What does each rule protect against?

Each rule exists because a specific failure is easy for an AI writer to make.

A made up example: the same email, broken and fixed

The email below is invented for this post. It is a first email to a head of operations at a regional food distributor.

Breaking the rules: "Hi Anna, I saw you just won a big retail contract, congratulations. We can cut your fulfilment cost by 30% and I can offer you a 20% discount if you sign by Friday. Best, Marco, Head of Sales, +00 000 000."

Count the breaks. The contract is an invented detail unless the lead data says it. The 30% is an invented number. The discount and deadline are a promised offer. The name, title and phone number duplicate the automatic signature. The comma splices read fine, but the invented facts are the damage.

Following the rules: "Hi Anna, we help food distributors answer delivery and quote questions without a queue. Is the speed of your quote replies something you are looking at this quarter? If not, tell me and I will not write again."

Every claim in the second email is either about the sender or a question. It states nothing that the lead data cannot support, and it gives the recipient a clean way to say no.

How do you audit a sample?

A weekly check takes about half an hour and catches drift early.

1. Pull twenty sent emails at random, from more than one campaign.
2. Read each against the seven text rules and mark every break.
3. Open three threads where the lead said no or stop, and confirm the thread ended and the contact is on the suppression list.
4. Open three threads where the lead asked about price, and confirm either an approved number or a handoff.
5. Check one mailbox's sends against its daily limit.
6. Fix the knowledge or the rule that caused the break, not the single email.

A break that repeats is a rule or knowledge problem. A break that appears once is usually a data problem in one lead record.

What are we not saying?

Three limits, so the list is not oversold.

• Rules are not law. They help you stay inside the rules for cold email in the countries you send to, and they do not replace advice. See is cold email legal.
• Style guidance is separate. There is a set of writing style rules, such as length and the shape of the offer, that is not switched on for customers yet. We do not list it as a live rule.
• Verbatim templates bypass the writer. If a campaign uses a fixed template, the message is sent as written, so the rules above on text do not change it. The rules on suppression, limits and machine mail still apply around it.

How should you use this list?

1. Check your own rules against the seven. Does your agent or writer have any of them?
2. Add a "what the agent may state" list to the knowledge: prices, dates, claims.
3. Turn on the unsubscribe setting where your sending practice calls for it.
4. Test the stop path. Reply STOP to your own campaign and see what happens.
5. Read a sample of twenty emails a week and count the rule breaks.

For the wider public statement, see our outreach rules.